All SpatialBench grading is deterministic and answers are evaluated quantitatively against ground truths constructed from careful analysis. However, over time we found issues with some evals.

Problem Ambiguity

Some tasks might depend on specific analysis decisions not specified in the prompt. In these cases, missing or ambiguous task context makes it difficult to understand if incorrect answers are due to lack of capability or poor instructions.

Grading Threshold Sensitivity

Other tasks have numerical answers with tolerance thresholds that allow a range of valid analysis methods to pass. Sometimes these tolerances were inappropriately calibrated, e.g. set too narrow, rejecting valid analysis paths the domain expert did not consider when framing the problem.

Main Results

To gain confidence that SpatialBench tasks were adequately specified, we needed evidence that each expected answer could be reproduced by an independent expert attempt.

We constructed a human-verified subset of SpatialBench called SpatialBench Verified comprising 115 tasks. Each problem in SpatialBench Verified had at least one human expert independently reconstruct an answer that passes the task graders from just the task prompt and associated data.

We found the filtered subset preserved the relative ordering of model performance, but scores increased by 11.6 pp on average.

Example: Problem Ambiguity

This sa_02_microglia_oligo_inflammation_correlation task asked agents to partition a mixed gene list into microglial activation and oligodendrocyte inflammation signatures, score cells, find neighboring oligodendrocytes around each corpus callosum microglia using an “appropriate radius” and compute Spearman correlations at 90wk and 4wk.

With just the task and no solutions, a human reviewer returned:

  {
    "spearman_90wk": 0.14,
    "spearman_4wk": 0.29,
    "interpretation": 2
  }

The existing grader expected:

  {
    "spearman_90wk": 0.32,
    "spearman_4wk": -0.17,
    "interpretation": 2
  }

The reviewer actually agreed with the intended qualitative interpretation (‘2’): age-associated spatial coupling between activated microglia and inflamed oligodendrocytes. But their notes indicated the exact numerical values were not recoverable from the task as specified.

Why?

If you look at the original problem statement, open choices include:

• how to split the mixed gene list into two signatures

• whether shared inflammatory genes can appear in both signatures

• how to normalize the low-depth targeted MERFISH panel

• what radius counts as “neighboring”

• whether to pool sections or control within section

• whether the provided 50k subset has enough power for the 4wk correlation

A variety of defensible analysis methods were therefore possible given the task context, making it difficult to trust the eval result.

Example: Grading Threshold Sensitivity

This visium_bone_norm_within_niche_coexpression task asked agents to define bone-enriched Visium spots as the top quartile of COL1A1 (symbol for a collagen gene) counts, then compute the Pearson correlation between COL1A2 and SPARC within those spots. The desired biological conclusion was that COL1A2 and SPARC are strongly co-expressed in the bone compartment, consistent with a coordinated osteoblast matrix-secretion program.

Once again, without solutions and just the task, the human reviewer reproduced the intended biological conclusion and reported:

  {
    "n_bone_spots": 735,
    "col1a2_sparc_corr": 0.8514,
    "interpretation": 1
  }

The grader expected:

  {
    "n_bone_spots": 762,
    "col1a2_sparc_corr": 0.62,
    "interpretation": 1
  }

The reviewer’s notes actually found the correlation is highly normalization-dependent despite relatively stable biological interpretation. This is a clear example of a grader tolerance that was too narrow for the goal of the eval.

Methods

We conducted an initial round of review where 6 domain experts solved all 159 problems in SpatialBench at an average of 26 problems per expert. Human experts were given access to the task prompt and associated data, and were asked to produce an evaluation answer in the same format as the agent answer.

For each problem, experts were asked to share both the final answer they arrived at and a Python Jupyter notebook or R Markdown document using any bioinformatics libraries they thought appropriate. All answers were graded using the benchmark’s existing grader, with binary pass/fail scoring (all answer fields must be correct).

94/159 (59.1%) of problems passed the first round of review. Upon review of the failing cases we found that failures fell into several categories:

1. Poor solution effort

2. Valid effort but incorrect analysis

3. Unclear mismatch between human solution and ground truth

4. Ambiguous or weak eval specification

The first category of failures was mainly due to time constraints - over a week experts were asked to solve an average of 5-6 problems a day which proved to be unrealistic. To mitigate this, we carried out a second round of review.

For the second round of review we included all 65 problems that failed the first round as well as 10 passing problems to serve as a control. The passing problems were randomly sampled across the 5 different spatial technology types used in SpatialBench (2 per kit). Problems were assigned over a larger pool of 28 experts (3 problems per expert). Experts were given 2 days to produce solution artifacts for their assigned tasks.

Experts in the second round of review started afresh with just the task prompt and associated data. All problems were assigned to different experts between rounds.

Passing solutions were produced in Round 2 for 21/65 of the failing problems from Round 1. From the control set, passing solutions were produced for 8/10 problems in Round 2.

We took the union of problems that passed Round 1 review (94/159) and the 21 additional problems that passed Round 2 review, yielding 94 + 21 = 115 problems in SpatialBench Verified.

Supplements

• Subset of solution data

• Task assignment sheet

• Instruction document

Authors

Anirudh Narsipur, Deborah Hayoun, Ananya Gupta, Shivang Sharma, Benjamin Kesler, Zachary Hemminger, Sahar Nasr, Aashka Bhowmick, Sahiti Marella, Zhen Yang, Shon Kurian George, Soo Hee Lee, Qian Xu, Lior Schachaf, Harihara Muralidharan, David Calcagno, Birendra Saye, Asmita Lagwankar, Meenakshi Somadasan, Alex Urrutia, Ian Diks, Hannah Le, Kenny Workman

24 evaluations span primary tumors, organoids, xenograft models, lineage-tracing systems, and aging/intervention biology. The best agents score 11.1%.

Read the manuscript.
Interact with the leaderboard.

Benchmark Construction

Evaluations mix the assay types scientists use in practice.

A single task may depend on spatial transcriptomics, histology, single-cell references, and lineage-recording data. Solving them requires cross-assay reasoning, experimental-design awareness, and command of spatial workflows like tissue segmentation, niche analysis, and spatial differential expression.

This tests the transition from running data analysis to doing science.

Constructing ground truth in science

Deriving ground truth is very hard in long-horizon biology benchmarks. The same data can support multiple valid conclusions, and some published claims do not reproduce cleanly under unbiased reanalysis.

Candidate tasks are hardened through independent reproduction, randomized expert review, and trajectories from multiple model families.

Pairing deterministic grading with diagnostic rubrics

Grading uses deterministic functions over structured final answers. We grade recovery of scientific conclusions expressed through controlled biological vocabularies instead of numbers from individual statistical operations.

Results

Across 15 model-harness pairs and 1,080 trajectories, Gemini 3.5 Flash / Pi, GPT-5.5 / Pi, and GPT-5.5 / OpenAI Codex each passed 8/72 attempts (11.11%), with Claude Opus 4.6 / Claude Code close behind at 7/72 attempts (9.72%).

Judge rubrics over analysis ‘chokepoints’ provide interpretive tools

However, final-answer grading provides sparse diagnostic signal for long-horizon tasks. Chokepoint rubrics - analysis decisions expected to remain stable across plausible solution paths - are graded by judges as a companion diagnostic.

But rubric judging is probably not a replacement for verifiable grading

We were interested in exploring the utility of rubric grading for intermediate reward and asked if denser signal was partially aligned with endpoint quality. We conclude rubric scores are promising auxiliary tools, not substitutes for verifiable endpoint grading.

Rubric judging is reasonably stable over judge and agent model variation

We also stratified rubric scores by the source agent model whose trajectory was being judged to see if rubric patterns are consistent across judges. All four judges preserved the same broad ordering with some variability.

Actually looking at your trajectories will continue to be important

Pairing manual trajectory review with rubric and verifiable scores provides more tools to interpret model failures.

In practice, manual trajectory inspection is a first-class tool to understand this data. Eval authors maintain reproduction notes to provide a record for future benchmark updates, especially as stronger models may solve tasks through unanticipated but valid analysis paths that challenge current grading assumptions.

There is a path to agents that behave a lot like scientists

Results suggest compounding local analysis errors prevent reliable long-horizon scientific reasoning.

Before models can reliably reason about disease mechanisms, drug response, or other deep results in biology, they must become procedurally competent in local steps.

But the few task completions we observed were very impressive. It seems there is a realistic path to agents thinking and behaving a lot like scientists do.

Read the manuscript
Interact with the leaderboard

On accuracy, Gemini 3.1 doubles, Opus 4.7 shows modest improvement and GPT 5.5 shows no significant improvement.

On latency, GPT 5.5 improves greatly, Opus 4.7 shows no significant improvement and Gemini 3.1 worsens by ~5x.

Full data and example evals/trajectories available on benchmarks.bio.

Models are showing steady improvement in generic data analysis ability

Trajectory review reveals improvement in exploratory behavior:

• Thoroughly investigating data and thinking through experimental context before picking which analysis algorithm or tool to use

• Self-verification after finishing steps or submitting answers

• Early signs of assay-aware analysis steps (eg. QC thresholds) specific to kit/machine type instead of generic tutorial steps. Largely explains the jump in Opus 4.7 accuracy.

Models still struggle with biology-specific analysis details

Many remaining evals require knowledge of experimental design, kit/machine-specific analysis details, or tacit scientific knowledge. Some examples:

• Aggregating to the correct statistical unit before testing (ex: inappropriately treating cells as statistical replicates, inflating power in statistical tests)

• Applying whole transcriptome scRNA-seq defaults to targeted 30-400 gene panels (ex: BD Rhapsody)

• Accurately recognizing and removing depth/batch confounds (ex: PC1 absorbs library size)

• Integrating data from multiple platforms (ex: DNA + RNA on Tapestri, protein + mRNA)

Explore data and trajectories on benchmarks.bio

We regularly update our benchmark family with new models: benchmarks.bio. Encourage those interested in understanding what these benchmarks actually measure to inspect sample tasks and trajectories.

Introduction

It’s important to remember just how surprising the rise of software engineering agents was. A year or two ago, most professional engineers thought of AI tools as code generation assistants - useful for formatting, refactoring, or scaffolding unit tests - but never capable of grokking complex requirements, modifying large codebases, or making systems-level engineering decisions.

Today, 4% of all public GitHub commits are authored by Claude Code and Mythos is uncovering thousands of day-zero vulnerabilities *autonomously* in workhorse software. Agents build browsers and compilers. Many of my most skeptical peers now use multiple agents in parallel to tackle daily, high stakes work.

In hindsight, many people treat this capability growth as inevitable. Software, after all, is “obviously verifiable”: unlike other domains grounded in the messy, physical world, code runs deterministically on computers. But software is more than code and frontier agents are starting to show higher-level engineering judgment on open ended tasks.

I believe biology will follow a similar path. Research workflows in biology will reorganize around agents for the same reason software did: both domains contain a layer of work that is concrete, executable and feedback-rich. In software, that layer is code. In biology, its measurement-grounded data analysis.

From this analogy, we’ll build up three claims using real examples:

First, the first useful biology agents will be data-analysis agents, not autonomous scientists, just as coding assistants came before increasingly autonomous engineering agents.

Second, assay-specific data analysis is needed for true biological reasoning to emerge.

Third, the analysis layer becomes economically more important as molecular data generation grows.

Software fell first because code is executable and feedback-rich

To understand what I mean by “both domains have a layer of work that looks the same”, let’s first dig into what that means for software.

Software is written in code, which has some nice properties. You can run it quickly through a compiler and get outputs and errors. You can easily inspect the intermediate state of the digital objects you build up in a program by printing values or writing them to files.

Code is then an ideal substrate for agents to learn tasks. They can write and run chunks, inspect outputs and loop. But the useful things we build in software are far more than just code. They incorporate subjective and somewhat open-ended engineering decisions, many inherited from the domain in which they’re used and the problems they solve.

It is then surprising agents developed the capability to autonomously construct, for example, browsers. “The verifiability of code” does not explain the progression from passing tests to making nontrivial systems decisions about concurrency, caching, latency and security.

The obvious reason is skill in higher-order concepts emerged after focused training on the procedural skill of code is because you have to get to the parts of the problem where this knowledge is useful. It’s difficult to learn how to rate limit an endpoint if you struggle with Python function syntax or are unable to reason correctly about event loops.

But more subtly, and important to the analogy with biology, is detailed knowledge of your building material informs the higher-level decisions. It’s difficult to make good browser-level decisions (caching, rendering) without understanding the lower-level behavior of code.

Biology has a similar analytical substrate

So what does any of this have to do with biology?

Most modern biology papers are centered around large-scale measurement experiments and share the following structure:

1. Choose a biological model (cell line, animal, organoid) and variables/controls

2. Generate data from the model

3. Process the data

4. Creatively thinking about the analytic results in context of prior literature

5. Make scientific claim

If you trace how an agent might work through this flow to tackle research autonomously, I argue:

1. Analysis of assay-specific data analysis is necessary to reach stages where scientific reasoning is useful

2. Scientific reasoning tasks are shaped like high level software engineering tasks, both in complexity and their dependence on underlying procedural skill in data analysis for accuracy.

So the existence of a thick underlying layer of molecular data analysis scaffolds scientific thinking in a similar way that code scaffolds more complex software engineering judgement.

The best way to make this concrete is with an example.

An example: reprogramming fibroblasts into brain organoids

Let’s take a look at a recent study from Gordon et al. where authors harvest fibroblasts (very common ‘connective’ cells) from 55 autism patients, reprogram them into stem cells with Yamanaka factors and turn each stem cell into a brain organoid. Their central question is “do the different mutations in each patient affect the development of autism”.

This is a good example of the type of project (structure, scale, difficulty) we might want an agent to tackle in the future.

We can immediately see the question admits no “global verifiability”. It’s unclear how to produce a correct value to grade against (or what type of value would even make sense). A ground truth would depend on dozens of intermediate observations, some without clear scientific consensus: how organoids are quality-controlled, how cell types are annotated, how gene expression modules are defined.

But we can decompose the paper into smaller steps with clearer edges, some even with “locally verifiable” results.

Cell line QC

First, the agent compares mutations in whole-genome sequencing and RNA-seq data to quality control stem cells.

We grade the agent on which cell lines it drops and which mutations it recovers for each line.

Differential expression

Then, the agent looks across cell lines to recover genes that shift the most over time (as the cells develop).

We grade the agent on the agreement with the genes recovered by the author.

Perturb-seq

We then use CRISPRi to actually “knock down” these genes in the lab. The agent analyzes this Perturb-seq data to identify which of these genes had a functional effect when it was repressed.

We grade the agent on the genes the authors concluded were important.

Each of these analysis steps start from raw measurement data and build up to a well-defined scientific outcome. All of them are necessary for a scientist to draw a conclusion about the overarching question.

Data analysis is the gateway to scientific reasoning

As shown above, agents can decompose biological research into concrete analysis steps, with each step building upon the previous one. Let’s now get a sense of the higher-level tasks remaining after the agent works through the data:

• Compare data from different assay types to validate results: bulk + sc-RNA-seq, IP-MS + Perturb-seq

• Compare expression across technical (sequencing lane, differentiation batch) and biological (donor, cell line) covariates

• Scour existing literature for prior biological evidence (eg. known disease associations or functions of candidate genes)

These tasks are much more “AI Scientist” shaped. They require synthesis across modalities, judgment under uncertainty, and knowledge of prior biology. But they are not obviously more complex than the kinds of open-ended engineering judgment agents are beginning to perform in software. The practical bar is matching the judgment of a strong scientist: reasoning from the available measurements, controls, analysis choices, and literature to the conclusion best supported by the evidence.

And because this biological judgement is empirical, the agent must reason from noisy data and understand analysis assumptions well. Developing scientific judgement then requires mastery of the analysis steps and cannot be separated from it.

For instance, an agent would need to understand the statistical choices made in the differential expression analysis. Decisions like which samples are included, how genes are filtered, how covariates (like sequencing lanes) are factored out can modify the results. An agent could conclude “a gene is downregulated” because these genes genuinely co-vary across mutation or because the gene is capturing a hidden variable in the data: developmental maturity, cell-type composition, sequencing depth, some temperature in the lab, etc.

Another example is how metadata is constructed. If the agent is looking for genes that regulate development, the exact way it constructed the cell type labels in the data need to be understood. Depending on how we define “cells in early development”, low count genes in this population could surface from genuine biology or an incorrect definition of this population with subjective marker genes.

An agent with an understanding of these assumptions is equipped to make accurate scientific decisions. If it did the analysis itself, it also has these details fresh in context.

Where the analogy breaks

Biology is among the last frontiers of the empirical sciences. The ‘objects’ of study (living organisms) span angstrom (protein) to micron (cell) to meter (leg) scale, synthesizing phenomena across chemistry and physics. The measurements we use to study them are produced by instruments and protocols on the frontier of human knowledge with noise and error.

This complexity means the analogy to software breaks in at least two places.

First, biological ground truth is more difficult to define. Even apparently well-scoped tasks like clustering cells, annotating cell types or calling differentially expressed genes rarely have one canonical answer. This is an active area of research, with ongoing work to develop practical methods and tools to constrain ambiguity and model scientific intent verifiably. It remains very difficult.

Second, feedback is much weaker once we leave the analysis layer. For data analysis, the feedback loop still looks somewhat like software: run code, inspect outputs, loop. However, true feedback for higher order scientific reasoning requires some control over the data-generating process. We will likely make a lot of progress using ground truths constructed and graded a lot like code without control over this loop. But true autonomous science requires direct feedback with the substrate being studied.

This layer becomes more important as biology scales

Agents will need to develop data analysis capability first (at least for real work) and accurate scientific reasoning seems in part dependent on a detailed understanding of analysis. It’s important to now understand how much more important analysis might be in the context of data generation trends.

Molecular data generation follows an exponential curve and analysis costs are rapidly becoming more expensive than reagents/prep labor with new assay generations. An increasing proportion of biological work now happens after the experiment has finished in the lab.

If you play out these trends, the bottleneck will shift from producing measurements to interpreting them. The question is not if but when. Future datasets will dwarf the stock of past biological data, and human capacity to analyze them will not scale at the same rate. The answer to this will be agents spinning in harnesses that reliably take data to scientific conclusions, deployed in R&D workflows across biotech where data is bottlenecked today.

The surface area of computational biology is growing

People often equate ‘computational biology’ with narrow analysis tools, like sequence assembly or differential expression software. While these have been important contributions of the field, computational biology should really be thought of as “the broad use of computers to solve problems in biology”.

For years, the components have been building for a computational shift in biology: rapidly compounding molecular data, increasingly automated labs, and iterative design-build-test-learn workflows. Yet these advances have not translated into an obvious step-change in aggregate R&D productivity.

That now appears to be changing. The industry is reorganizing around AI and computing, in how experiments are designed, data is analyzed, companies are formed, and pharma structures partnerships. Computers are becoming the interface through which agents “see”, understand and eventually manipulate living systems.

The first useful deployment of these agents will not be autonomous AI scientists but analysis collaborators embedded in real workflows. This is both where true need lies and what they will get good at first. Biology has no shortage of interesting new ideas. The bulk of economically valuable agentic work is concentrated where computers can make progress on R&D cycles, augmenting scientists where they are today.

Thank you to Aidan Abdulali, Anirudh Narsipur, Brandon White, Kyle Giffin, Alfredo Andere, Aashay Sanghvi, David Yang, Hannah Le, Lada Nuzhna, Tim Proctor for revising and/or contributing to the thinking in this essay

GPT-5.5 nearly halves runtime relative to GPT-5.4, yet accuracy remains flat: 57.6% versus 57.4%. Opus 4.7 is similarly tied with Opus 4.6: 52.4% versus 52.8%.

Scientist-reviewed trajectories reveal gaps in assay-aware biological judgment: statistical design, spatial units, batch structure, and scientific interpretation.

Complete benchmark data and select trajectories are available on benchmarks.bio.

What SpatialBench measures

Spatial biology is a powerful measurement tool and an important category for agentic capability. Analysis workflows require a combination of code and biological reasoning: agents must handle large data, understand platform-specific details, contextualize with scientific goals and return a quantitative result that matches what a diligent scientist would compute.

SpatialBench measures this work: 159 spatial biology analysis tasks across platforms like Xenium, Visium FFPE, MERFISH, TakaraBio Seeker, and AtlasXomics DBiT-seq. Each task starts from a real analysis state and asks an agent to recover a specific biological result. The grader checks structured outputs against expert-derived references (a subset of examples available here).

While frontier models show an improvement in speed and step count, they do not improve in overall accuracy on this benchmark.

GPT-5.5 is substantially faster than GPT-5.4, cutting mean duration roughly in half and using far fewer steps. But its accuracy is effectively unchanged: 57.65% versus 57.44%. Opus 4.7 is also effectively tied with Opus 4.6: 52.41% versus 52.83%.

Accuracy by platform

Platform-level grouping shows GPT-5.5 improves on Visium, Xenium, and MERFISH, but trails GPT-5.4 on TakaraBio and AtlasXomics.

Similarly, Opus 4.7 leads Opus 4.6 on Xenium by 11.1 percentage points, ties on TakaraBio, and trails on Visium, MERFISH, and AtlasXomics:

A look at manually annotated trajectories

Trajectory review identifies recurring failure categories across model families:

• Treating cells, beads, spots, or barcodes as independent observations when the biological replicate is the donor, animal, tissue section, or timepoint

• Applying scRNA-seq normalization defaults to spatial platforms where they are inappropriate

• Clustering multi-sample data without integration, then interpreting donor or timepoint structure incorrectly

• Confusing spatial marker units for cells or anatomical structures

• Failing to recover correct de novo niches, tissue states, and spatial compartments

We’ll dive into some examples to get a sense of failure modes in the context of realistic tasks. Each task trajectory was reviewed by a scientist with years of experience on the specific spatial platform being evaluated.

Barcode-level statistics create false positives

The AtlasXomics SPATIAL10_genome_wide_de_pct task asks the model to test 24,919 genes for sex differences in a human dorsal root ganglion DBiT-seq dataset. The dataset contains about 10,000 spatial barcodes from 8 donors: 3 female and 5 male.

Barcodes are nested within donors. A researcher would aggregate at the donor level, computing about 1.2% differentially expressed genes. The agents consistently ignore this donor information.

GPT-5.4 and GPT-5.5 both report 93.876% on all six runs. Both Opus 4.7 and Opus 4.6 report roughly 92-94% of all genes as significantly sex-different. Note this interpretation is biologically implausible: sex cannot plausibly alter chromatin accessibility at 93% of all genes in 8 donors.

Housekeeping genes expose the same pseudoreplication error

The SPATIAL07_sex_housekeeping_de task asks whether 10 housekeeping genes show sex differences in the same spatial ATAC-seq design. The expected answer is none.

Trajectory review reveals models call 9-10 housekeeping genes significant. Both Opus 4.7 and Opus 4.6 models call all 10 housekeeping genes sex-different in every run. GPT-5.4 and GPT-5.5 call 9-10 significant housekeeping genes. Calling ACTB, GAPDH, and other constitutive genes sex-differential is a clear sign of pseudoreplication: the models treat thousands of barcodes as independent replicates, artificially inflating statistical power and manufacturing significance.

Targeted-panel normalization flips myelin coexpression

The MERFISH norm_02_myelin_gene_coexpression_normalization task asks for the Spearman correlation between Mbp and Plp1 in oligodendrocytes. These are myelin structural genes and should be positively co-expressed. The expected value is about 0.308.

GPT-5.5 fails all three runs, consistently producing -0.157. Human review shows GPT-5.5 applies a normalization step that over-corrects targeted-panel counts, turning a positive biological relationship into an apparent anti-correlation. GPT-5.4 passes 2 of 3 runs with correlations around 0.326 by avoiding that specific normalization path.

Both Claude models report a Spearman correlation around -0.16 between Mbp and Plp1 on every run. In this case, the negative correlation is an artifact of library-size normalization on a 374-gene targeted panel where a few myelin genes dominate total counts. The models import the familiar scRNA-seq code:

sc.pp.normalize_total(adata, target_sum=1e4)

sc.pp.log1p(adata)

on a targeted MERFISH panel, rather than handling the platform appropriately.

Batch-confounded clustering survives newer models

The batch_driven_clustering (TakaraBio) and NORM01_batch_correction (AtlasXomics) tasks test whether models integrate across donors or timepoints before interpreting clusters.

Both GPT-5.4 and GPT-5.5 fail TakaraBio batch_driven_clustering. The expected max single-timepoint fraction is about 0.375; GPT-5.4 reports 0.967, 0.995, and 0.995, while GPT-5.5 reports 0.990, 0.994, and 0.988. Each cluster is dominated by a single timepoint, meaning the clustering is tracking condition or batch rather than cell type.

The same issue appears in AtlasXomics. In NORM01_batch_correction, the expected mean max sample fraction is 0.375, but the GPT computed values remain around 0.866-0.897. The Opus 4.7 trajectories independently describe the same AtlasXomics failure: without integration, PCA captures inter-sample technical variance, Leiden clustering partitions cells by sample of origin, and neither model questions why clusters in a multi-donor dataset are dominated by single donors.

Spatial units are usually not cells

TakaraBio Seeker uses 10um beads. A single large oocyte can span many beads, with RNA diffusing across all of them. Counting marker-positive beads as cells or anatomical structures inflates biological counts.

The oocyte_count_per_timepoint task tests this. The expected immature oocyte count is 850. GPT-5.4 reports 1237-2086, and GPT-5.5 reports 1510-3463. At the 0h timepoint, expected count is 275, while models report 424-821.

The same report highlights cumulus_gc_count_immature: the expected cumulus granulosa cell count in immature samples is 0 because cumulus cells have not differentiated yet. GPT-5.4 reports 435-1474, and GPT-5.5 reports 1424-2395. Both models assign cumulus identity from marker expression without applying the developmental constraint that cumulus differentiation requires hCG stimulation.

The Opus 4.7 report describes the same class of failure in spatial segmentation terms. On follicle_count_immature, Opus 4.7 counts span 50 to 456 across runs because small changes in DBSCAN radius produce wildly different connected components, while Opus 4.6’s stricter oocyte scoring gives a cleaner input set for spatial clustering.

Niches are hard to predict

The Xenium spatial_fibro_inflammatory_niche_emergence_2 task asks the agent to reconstruct a fibro-inflammatory niche across kidney injury timepoints. The expected pattern is low sham co-localization, high Day14 co-localization, and a fold increase of 6.9.

GPT-5.4 reports sham co-localization of 0.112-0.124 and fold increase of 2.26-2.32; GPT-5.5 reports sham co-localization of 0.121-0.373 and fold increase of 1.99-4.23. The expected sham value is 0.033, expected Day14 is 0.23, and expected fold increase is 6.9.

Opus 4.6 and Opus 4.7 find niches containing fibroblasts and immune cells, but they cannot enforce the specific composition ratios that distinguish the pathological CN7-like niche from adjacent healthy stroma. They confuse generic fibroblast-immune proximity with the disease-organized fibro-inflammatory compartment.

More development of common failure behaviors

Replicate-aware statistics: Models often misunderstand how to define replicates. When donor, animal, section, or timepoint is the appropriate variable of comparison, models incorrectly reach for cells, beads, and barcodes as independent observations, inflating significance and creating false positives.

Platform-aware normalization: Models treat spatial assays as interchangeable and often reach for scRNA-seq defaults. The same normalization procedure can overcorrect targeted MERFISH/Xenium panels, confound sparse Visium FFPE correlations with depth, distort bead-capture signal, or discard curated marker genes through HVG selection.

Batch and donor integration: Models frequently cluster before asking what the axes of variation represent. In multi-donor or multi-timepoint data, uncorrected PCA/Leiden structure often separates samples, batches, or timepoints; the model then mistakes this separation for cell type, tissue state, or treatment biology.

Spatial unit and denominator errors: Models confuse expression and coordinate units with cells or other structures.

De novo spatial niche discovery: Models can assemble the mechanics of niche analysis but often miss the biological goal. They confuse generic proximity or broad regional enrichment for specific compartment the task asks for: a follicle, lineage, pathological niche, or disease-organized tissue state.

Assay-specific benchmarks are important

While GPT-5.5 and Opus 4.7 are faster or locally stronger in specific task families, they are not meaningfully more reliable on SpatialBench.

Future improvements for spatial biology are unlikely to come from general reasoning gains alone and will require explicit training on statistical design, platform-specific analysis steps, replicate-aware differential testing and other spatial biology knowledge. Focused, assay-specific benchmarks are required to appropriately measure the complexity of biological data analysis accurately.

Here we introduce scBench (read the paper), a benchmark of 394 verifiable problems drawn from real scRNA-seq workflows. It spans six sequencing platforms, seven task categories, and eight frontier models. The best model reaches 52.8% accuracy. Better than spatial, but it means the best agent in the world still fails nearly half the time on routine analysis tasks.

scBench tests what bioinformaticians actually do

Existing biology benchmarks reward textbook recall or literature-style reasoning. Real analysis requires loading messy datasets, writing code, making judgment calls about thresholds and parameters, and producing concrete quantitative results. scBench recreates these conditions.

The benchmark covers 394 problems across six sequencing platforms (Chromium, BD Rhapsody, CSGenetics, Illumina, MissionBio, ParseBio) and seven task categories (QC, normalization, dimensionality reduction, clustering, cell typing, differential expression, and trajectory analysis). Each problem provides a data snapshot - typically an AnnData `.h5ad` file-captured right before a decision point, a natural-language task prompt, and a deterministic grader that scores the agent’s structured JSON output as pass or fail.

We pressure-tested each problem for shortcuts: removing precomputed embeddings, stripping cached labels, and ensuring answers can’t be guessed from prior knowledge alone. Cell typing (118 problems, 30%) and differential expression (71, 18%) dominate the benchmark because these are the stages where dataset-specific judgment matters most and where agents struggle hardest.

Agents are better at scRNA-seq than spatial - but still not reliable

Across eight frontier models from four providers, accuracy ranges from 29% to 53%.

Claude Opus 4.6 leads at 52.8%, followed by Opus 4.5 (49.9%), GPT-5.2 (45.2%), and Sonnet 4.5 (44.2%). The bottom tier - GPT-5.1 (37.9%), Grok-4.1 (35.6%), Grok-4 (33.9%), and Gemini 2.5 Pro (29.2%) - trails by a wide margin. The 23.6 percentage point spread between best and worst exceeds SpatialBench’s 18.3 point spread, meaning scBench discriminates model capability even at higher overall accuracy.

The accuracy bump over SpatialBench likely reflects training data. scRNA-seq has far more public datasets and Scanpy dominates the ecosystem with extensive documentation and tutorials.

Tasks reveal a consistent difficulty gradient

Not all analysis steps are created equal. Normalization is easiest (cross-model mean 70.4%), followed by QC (55.3%). These are procedural---apply a well-known transformation, check a metric against a threshold. Agents handle them reasonably well.

The story changes for judgment-heavy tasks. Clustering drops to 38.3%, cell typing to 34.9%, and differential expression to 27.0%. Seven of eight models follow this same difficulty ordering. DE is also where models diverge most: 27.7 percentage points separate best from worst.

The pattern is clear. Tasks that require contextual scientific reasoning-selecting marker genes, interpreting cluster identity, choosing statistical tests, identifying tissue-specific signatures-are where agents break down. General-purpose coding skill is necessary but not sufficient.

Platform choice matters as much as model choice

This is perhaps the most striking finding. Cross-model mean accuracy ranges from 59.1% on CSGenetics to 26.4% on MissionBio---a 32.7 point gap that exceeds the 23.6 point spread between best and worst models. Platform matters more than which model you use.

CSGenetics is easiest for six of eight models. MissionBio is hardest for all eight. The MissionBio collapse is dramatic: even the best model (Opus 4.6) only hits 42%, and Gemini drops to 10.3%. Every model shows large platform swings---Gemini drops 42 points between its best and worst platform. Even Opus 4.5, the most consistent model, loses 39 points.

MissionBio inverts the overall rankings in interesting ways. Grok-4 (sixth overall) beats GPT-5.2 (third overall) on MissionBio. Sonnet 4.5 surpasses GPT-5.2 by 11 points. Models that memorized Scanpy tutorials without learning transferable analysis techniques collapse on platforms with non-standard data structures and less common tooling.

These effects almost certainly reflect training data composition. Chromium and Illumina dominate public repositories and documentation. MissionBio and ParseBio appear less frequently. Reliable agents will need platform-aware context and assay-specific tooling, not one-size-fits-all reasoning.

scBench complements SpatialBench across single-cell modalities

Together, scBench and SpatialBench cover the two dominant transcriptional assays. The top model reaches 52.8% on scBench versus 38.4% on SpatialBench-scRNA-seq is more tractable. But the structural patterns are shared: normalization is easiest in both, platform effects drive 30--40 point swings in both, and model rankings are preserved at the extremes (Claude Opus leads both, Gemini ranks last in both).

The benchmarks form a complementary diagnostic. scBench tests whether models can handle the most common and best-documented assay type. SpatialBench tests whether that competence transfers to newer, less standardized technologies. Together they reveal whether an agent has learned general analysis reasoning or just memorized Scanpy workflows.

Limitations

Deterministic grading enables verifiable evaluation but necessarily discretizes scientific judgment into automatically checkable chunks. Each evaluation snapshots a single workflow step rather than capturing the long-horizon iteration where errors compound and thresholds are revisited. Real analysis is messier, more iterative, and longer than any single problem captures. But measuring step-level competence reliably is a prerequisite for automating longer workflows.

Where this is going

scBench confirms the pattern SpatialBench established: agents for biology are in a capability regime where they can accelerate routine analysis but cannot yet be trusted to autonomously answer scientific questions without human oversight. The path forward is a long tail of tractable engineering: platform-specific context, better harness design, and exposure to representative workflows across diverse biological contexts.

We’re building toward a family of benchmarks spanning major biological modalities, each an evolving formalization of the tacit knowledge and judgment that working scientists bring to data analysis. The goal is test-driven development of agent systems that improve through both model training and harness engineering.

Code, canonical evaluations, and full trajectories are available at github.com/latchbio/scbench.

This gap is starting to matter. Spatial assays (and increasingly every modern measurement modality) generate outputs that are too large, too irregular, and too context-dependent for most labs to analyze without computational specialists. The promise of agents is clear: point them at raw data, ask scientific questions in plain language, and compress weeks of analysis into hours. But looking right isn’t the same as being right - especially when results guide expensive and time intensive experiments.

Software engineering arguably went through this phase already. Frameworks like SWE-bench exposed how bad models were at real programming tasks and gave everyone a number to climb. Progress in biology needs something similar: built from real workflows, grounded in real datasets, and graded by verifiable outcomes instead of vibes.

Here we introduce SpatialBench (read the paper), a suite of spatial biology analysis problems designed to recreate the work scientists actually do - inspect messy images and count matrices, write and run code, reason scientifically and recover concrete biological results.

SpatialBench approximates real analysis work

Current biology evaluations largely reward textbook knowledge, but real analysis involves empirical interaction with data to produce answers. SpatialBench is a suite of 146 verifiable problems drawn from real spatial analysis workflows, spanning five platforms (Vizgen MERFISH, Takara Seeker, 10x Visium, 10x Xenium, Atlasxomics DBIT-seq) and seven task categories (QC, normalization, dimensionality reduction, clustering, cell typing, differential expression, and spatial analysis).

Each problem is a snapshot of an analysis workspace captured right before a decision point: typically an AnnData object plus any associated files, images, or outputs from earlier steps. The tests probe for durable biology, robust to survive reasonable choices of algorithm, library, or hyperparameters. We pressure-tested each for shortcuts: inspecting logs, adversarial prompts and removing anything solvable by pattern-matching or leaning on general scientific knowledge instead of actual analysis. Find concrete examples in the paper supplements.

Agents aren’t quite there yet

Across the full benchmark, base-model accuracy sits between 20-38% depending on the model family, with the usual tradeoffs between accuracy, latency, and cost.

But the aggregate numbers hide detail. Stratification reveals clear model-specific variation over tasks and platforms. Some model families collapse on specific task types while others swing 15 to 20 points depending on the class of spatial technology.

Spatial biology is of course a collection of many distinct technology types and constituent steps, each with its own artifacts and conventions obscured in point estimates of accuracy. Deliberate engineering will be needed to improve across each covariate.

Harnesses are more than “glue code”

Interestingly, harness design moved outcomes as much as swapping the base model. We ran Opus-4.5 under multiple execution ‘wrappers’ - LatchBio, Claude Code and a slightly modified mini-swe-agent - and saw dramatic variation.

Our results suggest we should start treating harnesses as first class objects in engineering and benchmarking efforts. The tools, prompts, and control flow around a model are as important as the model itself.

Agent trajectories reveal distinct problem solving strategies

Manual inspection of agent trajectories - the sequence of reasoning traces, tool calls, logs from an eval session - was perhaps the most informative part of the project. This data provides mechanistic explanations for performance differences. We could pick out distinct failure modes: models burning steps on formatting with no forward progress; using general QC thresholds that don’t hold for spatial data. Some models were able to ‘productively explore’ (inspect, compute, refine, make some forward progress) while others thrashed (same failed attempt on repeat, often forever). There are many examples of these in the paper.

Limitations

Verifiability comes with tradeoffs. Nuance is lost trying to draw deterministic lines around messy real world examples. Snapshots of single steps aren’t full workflows: real analysis is longer-horizon and iterative. But measuring step-level chunks with reliability is but a step towards automating truly long time horizon tasks. More on this soon.

Where this is all going

SpatialBench is a first contribution towards a broader family of benchmarks spanning major biological modalities. We view benchmarks as evolving formalizations of tacit scientist behavior and analysis workflows. They guide test-driven development of agent systems through model training and harness engineering that will eventually automate much of what we consider to be computational biology.

2026 will be a big year for agents in biology.

Read the paper. Check out data + code. Play with some examples.

Try it here: agent.bio

We’ll go over:

• Our roadmap for reliable and widely deployed spatial agents

• Important considerations while playing with these sandboxes

• Video demos of each agent

Our roadmap for reliable and widely deployed spatial agents

LatchBio develops agents for spatial biology data that allow scientists to direct their own analysis with natural language. By tailoring each agent to specific spatial kits and machines, and iterating relentlessly on this reduced scope, we believe it’s possible to build accurate systems that can be used to make expensive scientific decisions.

This release represents a shippable intermediary towards this goal, both to show technical progress and help others see why we’re so excited about this direction. The path ahead is a long tail of soluble engineering work: continued progress on benchmarks, purpose built infrastructure and technology specific heuristics encoded into prompts/evals.

Important considerations while playing with these sandboxes

We’re developing two agentic modes, “step-by-step” and “proactive”, that provide control over how often the agent will check-in, ask questions and provide figures or statistics for scientists drivers to sanity check intermediate steps. The sandboxes default to “proactive” to provide a nice first impression of end-to-end functionality, but we encourage biologists to start with “step-by-step” for any work of value. More on this soon.

Furthermore, the sandboxes were built for specific datasets and tasks. You can try to use your own data, and the system will likely generalize accurately on some tasks, but we provide no guarantees on performance quite yet.

Video Demos

The following videos and descriptions motivate some of the practical end-to-end workflows possible with the system. We highly encourage the reader to play with each themselves.

Takara Seeker

Ovulation failure affects over 10% of infertility cases in women of reproductive age in the United States, yet it is exceptionally difficult to study because ovulation unfolds within tiny, spatially restricted niches and evolves on the scale of minutes.

Using a Takara Seeker 3x3 kit, Mantri et. al built a spatiotemporal molecular atlas of the ovulating mouse ovary across 8 different time points.

To replicate the analysis, the Takara agent ingested raw H5AD files, removed off-tissue beads, computed QC metrics and filtering, performed clustering, and ran differential gene expression. The agent then manually reasoned through top marker genes for each cluster and identified cell types key to the ovulation process, such as mural and cumulus granulosa cells, mesenchymal subtypes, oocytes, and endothelial and immune populations. A scientists can then ask the agent to count the atretic or antral follicles.

Vizgen MERFISH

The mouse brain contains highly organized cellular circuits, yet these structures are difficult to resolve without single-molecule spatial data. MERFISH overcomes this barrier by capturing the precise locations of individual transcripts across intact tissue.

Using a MERFISH mouse brain dataset, the agent built a spatial neighbors graph, computed Moran’s I, and performed neighborhood enrichment and co-occurrence analyses. It identified spatially variable genes such as MYH11, SLC17A7, and DAAM2, each marking distinct anatomical regions; MYH11 was correctly enriched in blood vessel-associated cells, consistent with its vascular smooth muscle role. Neighborhood enrichment further recovered expected spatial organization, including co-localization of excitatory and GABAergic neurons and strong oligodendrocyte self-enrichment, matching known cortical and subcortical structure.

AtlasXomics DBiT-seq

Tumor heterogeneity is a major challenge in cancer, arising both between patients and within a single tumor. Changes in chromatin accessibility are a key driver of this variability because they alter gene expression, cell identity, and interactions in the tumor microenvironment.

Using AtlasXomics spatial ATAC-seq (DBiT-seq) from a stomach cancer sample and a matched control, the agent ran an ArchR-based workflow to identify differentially accessible chromatin regions, peaks, and motifs.

In tumor tissue, the agent detected strong enrichment of CTCF/CTCFL motifs, pointing to disrupted 3D chromatin organization, a hallmark of many cancers. It also found IRF1 enrichment, consistent with immune infiltration or localized inflammatory signaling. By contrast, the control tissue was enriched for FOXA1 and TEAD motifs, which support normal epithelial identity and Hippo pathway-mediated tissue homeostasis.

10X Xenium

Alzheimer’s disease reshapes the brain through localized inflammatory niches and region-specific neuronal vulnerability. These patterns only become visible only with high-resolution spatial transcriptomics.

Using a Xenium dataset from a TgCRND8 Alzheimer’s mouse brain section, the agent loaded the raw outputs, built an AnnData object, performed clustering and differential expression, and annotated cell types using a controlled mouse brain vocabulary.

The analysis recovered five major neural populations: astrocytes, excitatory neurons, inhibitory neurons, oligodendrocytes, and microglia. It also detected Alzheimer’s-relevant signals such as strong Gfap spatial enrichment marking reactive astrocytes and Hexb expression marking activated microglia. Finally, the workflow identified 21 spatial domains that correspond to known anatomical regions and disease-affected microenvironments, including glial clusters surrounding neurons, a characteristic pattern in Alzheimer’s models.

10X Visium

Colorectal cancer is shaped by tightly packed tumor, stromal, and immune niches that often sit only a few microns apart, making high-resolution spatial profiling essential for understanding disease architecture.

Using the Visium HD colorectal cancer dataset with more than 500,000 spots, the agent converted Space Ranger outputs into an H5AD and launched a GPU-accelerated RAPIDS workflow to handle the dataset at scale. The full pipeline from QC to clustering to differential expression completed in under a minute.

The agent then inspected top marker genes for each cluster and assigned epithelial, stromal, and immune cell types, reconstructing the organization of the colorectal tumor microenvironment.

Path Ahead

We’ve learned a lot about biology specific agent engineering in the past month building and saturating our benchmarks (eg. what is the analogue of a “code diff” for spatial analysis). Expect more educational material on these topics in the coming weeks.

High scores on SpatialBench represent expert and industrially relevant analysis ability across four different spatial technologies, tissue types and diseases. We benchmarked our spatial agent against Sonnet 4.5, GPT-5-Codex and Qwen 2.5 Coder in a minimal harness (a slightly adapted version of the Bash only SWE-bench harness). These early results show considerable engineering is needed to extend computer use agents to biology. There is a lot of work remaining to saturate these benchmarks.

We’re also releasing a batched test harness, eval spec and a representative subset of tasks/datasets on GitHub. While we built this tool to measure and improve our own product, we hope others find the open resources useful.

SpatialBench evals represent practical analysis tasks

This benchmark is focused on real world analysis work and each task was constructed actual chunk of work from our customers and their scientists.

Each eval is a JSON object conforming to a standard structure: a task description, data node and grader configuration. This information is injected into a system prompt and the agent performs work in a sandbox with access to data. A grader then evaluates the answer.

Here we demonstrate some concrete examples to build some intuition for what the benchmark is actually measuring. The specs for these evals are available in the repo.

Normalization

Feature Selection

Dimensionality Reduction

Clustering

Differential Gene Expression

Domain Detection

Results demonstrate need for biology specific agent engineering

While the benchmark does not represent an apples-to-apples comparison of systems, eg. a ~100 line mini-swe-agent harness wrapping Sonnet is significantly less complex than our system, the purpose of the exercise was to show what worked for coding will not necessarily work for biology. We will be improving the harness and expanding the eval bank over the coming weeks.

Putting a model in a simple loop with access to a shell is sufficient to saturate SWE-bench, but agents for spatial biology likely need purpose built infrastructure, tools and context for useful work. More on the details here.

Evals encode tacit knowledge of frontier research biology

The frontier of research biology - practical knowledge about details of different diseases and tissues - is distributed across hundreds of labs, in the minds of PIs and researchers. By working with these labs to interpret spatial data from different tech, we can encode their knowledge and preferences into evals used to train and benchmark agentic systems. These eval banks will become valuable, structured documentation about the procedural details of research as they grow.

Add your spatial technology to our benchmarks

We are expanding SpatialBench to include all relevant machine and kit types. Reach out to our team to include your technology in our next release.

Unlike most experiments in biology, the computational steps in spatial can match or exceed the resources spent on upstream work in the wet lab: a tricky combination of the data being large (>1TB a run for CosMX) but the analysis high-touch, multi-turn and very unique to each scientists’ questions. This is unfortunate because most biologists have a very clear idea of what they are trying to do, often able to describe the end results/plots with great detail and steer intermediate steps based on their biological goals. But the practical complexity of using spatial tools complicates the translation of the correct ideas to the right analysis. Tens of thousands of scientists across industry and academia are stuck analyzing spatial data .

Agents are LLMs in loops that plan and execute end-to-end analysis with natural language directions by interacting with their environment and calling tools. They have shown incredible promise in math and programming, but require intentional engineering to translate well to biological data. Latch spatial agents are tailored to the details of each kit, in direct collaboration with the companies that sell them, so they have the specific context, tools and structured analysis flows required for each type of data. They build scientific notebooks with interactive widgets running on large machines and launch workflows on purpose built data infrastructure engineered for scale.

We are in the early days of this emerging category, but the initial technical results and response from scientists indicate this is likely the future of data analysis. There is considerable work left to do, but it seems base reasoning and agentic capabilities of frontier models are essentially here, and the remaining structure of the problem is a long tail of incremental but very soluble engineering improvements. In the coming months, tens of thousands of scientists will be able to finally play with their own data and answer their own questions.

• Agents for biological data are purpose built for the domain

• Spatial data is a strong focus for early adoption

• What do agentic workflows look like in practice?

  • AtlasXOmics DBiT-seq Agent

  • Takara Trekker Agent

  • Vizgen MERSCOPE Agent

  • 10X Xenium Agent

Agents for biological data are purpose built for the domain

Agents need new capabilities to reliably and accurately accomplish practical tasks with biological data. Unlike software engineering products, these systems need to run in sandboxes with considerable computing and storage resources to manipulate large data. They also need to offload particularly intensive steps to optimized workflow infrastructure; single operations, like processing sequencing reads or clustering counts, can take hours and require clusters of computers. Using pre-existing bioinformatics components as tool calls, like Nextflow or Snakemake workflows, also serves to constrain end-to-end workflow variability and allows models to focus on genuinely novel components of user requests.

Furthermore, the driver of the agent is a scientist who reasons about results with plots, charts and graphical components. Latch agents communicate with their users by building scientific notebooks, rich with markdown description, that also serve as self-documenting dashboards that describe the analysis in detail.

Perhaps most important, each agent is tailored to the specific type of machine or kit that generated the data. We work directly with the companies developing these assays when possible - Vizgen, TakaraBio, AtlasXOmics - to encode latent knowledge about the specific details, hidden footguns and ideal customer journey into prompts, tools and evals. Rather than an open-ended sandbox that does many things somewhat well, our philosophy is to build a surgical and highly reliable product that assay providers can trust. We believe the bar for widespread adoption of agents is an intelligent, adaptable system that does not jettison the determinism of traditional software.

Spatial data is a strong focus for early adoption

It is important to find specific tasks that are uniquely suited to agentic workflows and cause enough existing pain to encourage scientists to try a new tool. Spatial data checks both of these boxes. This is a new, rapidly growing but notoriously challenging domain of science. There are no clear analysis standards, the ecosystem of tools is highly fragmented and generally difficult to use, and unlike traditional sequencing, there are many different, composable steps. Yet the power of these assays is remarkable, in the new biology they help us understand and the beautifully intuitive way they allow scientists to see what’s going on.

In many ways, it is convenient agentic systems are beginning to actually work. It is unclear exactly what the solution to these problems would have been with traditional software: likely many variations of hard coded dashboards that each tackle a specific task in a reliable but brittle manner. We believe the interplay between spatial biology, which is going through its own revolution, and agentic data analysis is a special, reciprocal relationship. These complementing technologies will allow scientists to ask more, understand more and all the while consume more kits and compute.

What do agentic workflows look like in practice?

The following are four concrete end-to-end agentic workflows built in collaboration with real scientists. A tailored agent is built for each kit or machine type. Scientists begin with raw data, send instructions or questions to a chat interface (eg. “Process my Vizgen data”), and watch the agent build graphical scientific notebooks to accomplish tasks.

AtlasXOmics DBiT-seq Agent

AtlasXomics develops an assay called deterministic barcoding in tissue sequencing (DBiT-seq) that allows spatially-resolved epigenome profiling. Microfluidic chips deliver barcodes to precise tissue coordinates, uncovering relationships between chromatin accessibility, histone modifications, and context-specific gene expression within intact tissues.

This first flow highlights the design principles discussed earlier: large agent sandboxes to handle the TBs of file data, the use of existing bioinformatics workflows developed by the AtlasXOmics team and specialized visualization components (H5, IGV viewer) required for this data type.

1/ Bioinformatics Workflows as Tool Calls

The AtlasXOmics agent begins by building interfaces to handle user input and dispatching bioinformatics workflows to purpose built data infrastructure. Reads are filtered, aligned to the reference genome, and converted into fragment files.

It understands the relevant files to pull from the output folder - filtered fragments, spatial position files, motif enrichment scores and gene activity scores - so scientists can immediately continue with downstream steps. Scientists can ask questions about workflow parameters at any point.

2/ Image Alignment

Linking molecular measurements to precise tissue locations requires careful alignment of sequencing data with high-resolution tissue images.

The agent guides scientists through this process using an interactive interface, providing instructions on how to perform alignment and suggesting adjustments when needed. The result is a spatially-aligned dataset where each molecular measurement is mapped to its exact tissue coordinate, creating the foundation for all downstream spatial analysis.

3/ Spatial Region Selection

Once the data is aligned, scientists need to define regions of interest based on tissue morphology or biological features. The agent enables researchers to use intuitive lasso-select tools to define custom regions

It can suggest regions based on expression patterns or guide selection based on natural language descriptions like “select the tumor border region.” These defined spatial regions are saved as metadata, ready for comparative analysis across different tissue areas.

4/ Differential Chromatin Analysis

Understanding regulatory differences between conditions, spatial niches, or cell types requires identifying differential chromatin accessibility regions and motifs. When scientists ask questions like “What chromatin regions are more accessible in the tumor core versus periphery?”

The agent launches the appropriate differential analysis workflow as another bioinformatics workflow, automatically navigating the Latch file system using the LPath API to find required inputs, and configures parameters based on the scientific question. The analysis produces differential peak results, motif enrichment comparisons, volcano plots, heatmaps, and gene regulatory network differences.

5/ Subclustering

Discovering subtle cell populations or states within major cell types often reveals distinct regulatory profiles crucial for understanding biology. When scientists want to explore specific cell populations in detail, the agent subsets the data based on user criteria and launches re-clustering workflows on the subset.

It then repeats annotation and differential analysis on subclusters while maintaining the full analysis provenance in the Latch Plots notebook environment. This iterative process yields refined cluster definitions and subcluster-specific regulatory signatures.

6/ Visualizing Chromatin Accessibility Tracks with IGV Browser

Understanding how chromatin accessibility differs between clusters requires examining the actual peaks and signal intensities at specific genomic loci. When scientists want to validate differential accessibility results or explore regulatory elements in detail, the agent launches an IGV browser session and automatically loads cluster-specific chromatin accessibility tracks as bigWig files, with each cluster displayed in distinct colors for easy comparison.

Scientists can navigate to genes of interest by simply asking “Show me chromatin accessibility at the IL2 locus across T cell clusters,” and the agent configures IGV to display the relevant tracks, adjusts the viewing window, and highlights differential peaks. The agent can also overlay additional tracks like gene annotations, known enhancers, or transcription factor binding sites, providing the full genomic context needed to interpret spatial epigenetic differences.

Takara Trekker Agent

TakaraBio offers two spatial transcriptomics platforms. This flow focuses on Trekker, a single nuclei spatial transcriptomics kit.

We show the agent’s ability to gather experimental context, launch chemistry-aware bioinformatics workflows, guide scientists through tissue image alignment, and discover insights through multi-sample differential gene expression analysis.

1/ Gather Experimental Context

The chemistry of each assay dictates the structure and content of its sequencing output and therefore the architecture of its downstream computational analysis. To enable robust interpretation, we codified assay-specific scientific logic into our agentic framework, allowing our Latch agent to reason about experimental design, platform chemistry, and expected data modalities.

The agent is trained to ask clarifying questions about the kit chemistry, sample type, and sequencing platform before constructing the appropriate workflow. This ensures that each dataset is processed using a computational path aligned to its biochemical origin, enabling accurate spatial reconstruction and downstream biological analysis.

2/ Chemistry-specific Bioinformatics Workflows

The workflow ingests paired FASTQ files alongside output from a single-cell platform (e.g., 10x or BD Rhapsody). Each read encodes a cell barcode, molecular identifier, and spatial barcode. The Takara spatial informatics pipeline identifies nuclei carrying valid spatial barcodes, reconstructs the spatial index, and generates a cell × gene matrix that integrates molecular expression with spatial localization.

When prompted with, “Process FASTQ files from a Takara Trekker experiment,” the agent dynamically generates a configuration interface to capture parameters such as the single-cell platform, kit version, and output format. It then invokes the appropriate Latch `w_workflow`, which executes on dedicated high-performance infrastructure to produce an H5AD object ready for visualization in the Latch plotting environment.

2/ Image Alignment and Tissue Visualization

Seeing gene expression overlaid on tissue structure is crucial for understanding spatial data. However, aligning H&E-stained images with spatial coordinates is challenging and often requires manual tweaking. The agent simplifies this using built-in image registration tools in the H5 viewer.

When prompted, the agent automatically focuses on the tissue region and overlays the H&E image with the spatial data. For ovarian tissue, this alignment reveals how gene expression maps to anatomical features: follicles appear as distinct circular structures, corpora lutea show as larger irregular regions, and the surrounding stroma displays its own expression patterns. This visual overlay lets researchers immediately connect molecular data with tissue architecture.

3/ Differential Gene Expression Across Follicular Subtypes

We also demonstrated the agent’s ability to conduct differential gene expression analyses across ovarian follicle subtypes—preantral, antral, and atretic—within Seeker-derived datasets. This analysis established a spatially resolved baseline for folliculogenesis, linking molecular identity to anatomical localization, revealing spatial gradients in follicle maturation, and identifying novel, spatially restricted gene markers critical to understanding ovarian development.

When extended to temporal dynamics, the agent identified stage-dependent changes in gene expression following hCG induction, including a decline in Fabp3 and Inhbb levels over time, marking the transition of mural granulosa cells from a proliferative, follicle-maintaining phenotype to a luteinizing, pre-ovulatory state.

Vizgen MERSCOPE Agent

Vizgen’s MERSCOPE platform uses multiplexed error-robust fluorescence in situ hybridization (MERFISH) to map hundreds to thousands of RNA species simultaneously at subcellular resolution. Each experiment generates terabytes of data: millions of transcript coordinates, multi-channel z-stack images, and complex metadata files.

We show the agent’s ability to process TB scale MERFISH human brain imaging data, launch GPU-accelerated cell segmentation workflows, enable interactive visualization of millions of transcripts in the browser, and perform spatially-aware differential expression analysis through natural language commands.

1/ Cell Segmentation

Cell segmentation determines how millions of detected transcripts get assigned to individual cells. MERSCOPE captures five imaging channels across seven z-planes: DAPI (nuclei), three Cellbound channels (membranes), and PolyT (cytoplasm). Creating accurate cell boundaries requires selecting the right channel combinations and configuring GPU-accelerated algorithms like CellPose or Baysor.

The agent automates this complexity by scanning imaging directories and generating configuration files tailored to each tissue type. For dense tumor regions, it prioritizes membrane channels; for sparse samples, it leverages Baysor’s transcript-based approach. It then launches a GPU-powered workflow on Latch infrastructure to process the segmentation.

The workflow outputs three files: (1) an AnnData object compatible with downstream tools like Scanpy and Squidpy, (2) PMTiles (a pyramidal format originally developed for web mapping that enabling smooth zooming and panning), and (3) a DuckDB database containing transcript coordinates and metadata, providing a lightweight SQL engine for querying millions of transcripts in real-time.

These files are displayed in the Latch H5 viewer, where scientists can seamlessly overlay cells, transcripts, and high-resolution morphology. They can also filter transcripts by gene, spatial region, expression level, or cell type while exploring tissue architecture at any zoom level.

If segmentation needs refinement, scientists can immediately instruct the agent to rerun with adjusted parameters, creating a rapid iteration cycle for optimal results.

A video of agent performing all analysis steps end-to-end for Vizgen data, from QC, normalization, finding HVGs, to cell type annotation and differential gene expression

2/ Quality Control

With cells defined, quality assessment becomes critical. MERSCOPE’s large imaging area can harbor technical artifacts: edge effects, tissue folds, or regions with poor probe penetration that must be identified before analysis. The agent generates spatial QC plots revealing these issues, allowing scientists to set informed filtering thresholds that preserve biological signals while removing technical noise.

3/ Clustering and Cell Type Annotation

MERFISH panels contain 200-1000 genes carefully selected to distinguish cell types in specific tissues. The agent exploits this focused design, performing clustering at resolutions that match expected cellular diversity. Initial annotation queries reference databases for marker patterns, then scientists refine through natural language: “split the myeloid cluster into monocytes and macrophages” or “identify proliferating cells across all types.” Spatial context often resolves ambiguous populations. Morphologically similar cells may occupy distinct tissue compartments that clarify their identity.

4/ Differential Expression Analysis

With cell types defined, scientists often ask what makes populations distinct or how they change across conditions. When comparing “tumor cells in the core versus edge,” it accounts for spatial autocorrelation and multiple testing across the focused gene set. Results highlight not just which genes differ, but their spatial patterns: gradients of metabolic genes from hypoxic cores to vascularized edges, or immune activation markers concentrated at tumor-stroma interfaces.

10X Xenium Agent

10X Genomics’ Xenium platform uses in situ hybridization to detect hundreds of RNA species at subcellular resolution. To demonstrate the agent’s capabilities, we’ll analyze an Alzheimer’s disease mouse model dataset tracking pathological changes across disease progression.

We show the agent’s ability to parse complex multi-timepoint Xenium datasets, refine cell segmentation in challenging brain tissue, detect spatial domains through BANKSY clustering, and calculate disease-specific gene signatures that reveal spatially-organized neuroinflammatory responses around amyloid plaques.

1/ Loading and Merging Xenium Data

The dataset compares wild-type and CRND8 transgenic mice at timepoints representing mild, moderate, and advanced amyloid deposition. Scientists attach Xenium outputs from coronal brain sections profiled with 347 genes: 248 from the Mouse Brain panel plus 99 custom genes targeting activated microglia, astrocytes, and plaque-inducible markers.

The agent parses the complex directory structure, constructs an AnnData object preserving spatial coordinates, and generates diagnostic plots.

2/ Cell Segmentation Refinement

Brain tissue presents unique segmentation challenges. Neurons range from tiny interneurons to large pyramidal cells spanning hundreds of microns. Around amyloid plaques, activated microglia cluster densely, creating cellular aggregates that standard segmentation often merges incorrectly.

When prompted to “improve segmentation around microglial clusters,” the agent launches a refined workflow. Cellpose first identifies nuclei using DAPI, providing reliable seeds even in crowded regions. Baysor then expands these boundaries guided by microglial transcripts like Iba1 and Cx3cr1. This approach captures the complex morphology of activated microglia wrapped around plaques–essential for accurately quantifying neuroinflammatory responses across disease stages.

3/ Spatial Domain Detection

With proper segmentation complete, the agent reveals tissue organization through BANKSY clustering.

4/ Cell Type Annotation

Standard cell typing identifies expected populations, but the key insights come from disease-specific signatures. When scientists prompt to “compute PIG signature scores within microglia and astrocytes,” the agent executes a targeted analysis of plaque-induced genes, a curated set of markers that distinguish reactive from homeostatic glia in Alzheimer’s pathology.

The agent first identifies microglia and astrocytes using canonical markers, then calculates PIG scores incorporating genes like Apoe, Cst7, and Axl for microglia, and Gfap, Vim, and Clu for astrocytes. These scores are mapped spatially, revealing striking patterns: microglia closest to plaques show the highest PIG scores, forming a reactive barrier, while astrocytes display elevated scores in broader halos around pathological sites.

Try it

Shoot us an email at contact[at]latch.bio to get onboarded. Lots of work to be done, but we have certainly seen early signs of the magic. We believe this will soon be the standard way biologists interact with data.

Every biotech executive has the same nightmare: a data breach halts your trial, exposes your critical IP, and scares off future investors. GDPR exists to prevent this nightmare; in short, it is Europe’s strict privacy rulebook (but, unlike most lab SOPs, this one comes with fines up to €20 million when things go wrong).1

Despite its reputation in biotech as a bureaucratic hurdle, GDPR reflects principles that not only protect data, but in practice, handle it to an end that supports the reproducibility and validation of experimental output for clinical translation.

For solution providers, GDPR’s value is clearer; it presents a commercial advantage and aligns your analysis with the same values that patients, clinicians, regulators, and software solutions already demand: accuracy, transparency, and accountability. It creates a common standard upon which geographically-diverse client data moves seamlessly from research to clinic, across international borders, and through tools without losing integrity along the way.

This post will apply key GDPR principles to the lab, operationalizing them with LatchBio’s approach to default-compliant product development.

First, an Overview of Compliance at LatchBio

LatchBio’s Global Compliance Program is designed to optimize the speed of your discovery without sacrificing its security. We work closely with both technical and regulatory partners to ensure the highest level of compliance and confidence for our customers:

• Drata provides 24/7 compliance monitoring visible to you,

• NAMSA advises on biotech-specific regulatory positioning,

• Tenax Solutions conducts annual penetration testing,

• MJD Advisors audits the integrity of our commitments each year, and

• GRCI Law employs our EU- and UK-based GDPR representatives and advises on international transfers.

Our aim is to meet every client’s regulatory requirements and work closely to provide regulatory support needed from a data processor hosting critical scientific infrastructure. To this end, we are compliant with a range of US and international frameworks, and quickly work with our customers when they need custom requirements or to comply with a different regulatory framework.

Latch is compliant with HIPAA, SOC 2, GxP, CLIA, CCPA/CPRA, and more. Visit our Trust Center to view our extensive security and compliance documentation.

Data Provenance and Reproducibility for Patient Rights

Across industries and countries, the public increasingly demands visibility into what companies know about them (e.g., requesting a copy of your personal data from Meta, or asking 23andMe to permanently delete your data). GDPR codified this cultural shift into law by giving patients clear rights to access, correct, or delete what companies know about them,2 including biotechs and solution providers.

GDPR and complementary frameworks have requirements that are satisfied by immutable logs of processing history (e.g., every workflow, data transfer, analysis step, or single data manipulation),3 with evidence of who did what and when. If done poorly, this could result in a disarray of paperwork and messy file systems (think graduate student’s desktop with folders inside folders, labeled “final_v7_actually_final).”

Latch’s architecture treats auditability and provenance as defaults. The full history of manipulation and record is available without manual reconstruction. The result is a system aligned with GDPR’s principles of accountability and transparency.4

The value of this requirement and solution can be interpreted beyond data subject rights. It allows scientific teams to reproduce an experiment exactly as it was run, returning to prior states and tracing how data evolved over time without guesswork to confirm the reliability of results.

Principle of Least Privilege in Modern Bioinformatics Management

Research data moves constantly, between collaborators, instruments, and workflows. Each transfer introduces potential for error: a file overwritten by someone who didn’t need editor access, a dataset manipulated without context, a result tied to the wrong version of an input. GDPR formalizes what lab operational excellence already demands — that access to data is deliberate, limited, and accountable.5

The principle of least privilege6 is, at its core, a safeguard against that kind of entropy. It requires admins to limit access to only what each user needs for their role and to keep detailed records of those interactions. In other words, it encodes good lab hygiene into law.

On Latch, this takes form through granular role-based access control (RBAC). Workspace owners define broad roles (viewer, editor, or admin) and then granularly customize permissions within those roles. Before high-impact actions like data export or deletion, user identity is re-verified7, and Latch engineers would have already been monitoring automatic signals for suspicious log-in behavior to report to admins.

The benefit isn’t just compliance; it’s coherence. Least-privilege access enforces experimental discipline: teams work from a single, trusted source, data integrity maintains intact, and the UI of each employee is specifically designated by the admin. What GDPR describes as “security of the processing” is, in practice, the same discipline required for reproducible, multi-stakeholder science.8

Commercial Advantage of GDPR Compliance for Biotech Solution Providers

In any pharma sale, the client’s procurement team will ask for a key set of GDPR-required artifacts, to size up a vendor’s readiness for regulated data transfer:

• Standard Contractual Clauses (SCCs) define the terms for international data transfers9,

• Data Processing Agreement (DPA) outlines how a processor manages and protects a controller’s data10,

• Record of Processing Activities (ROPA) documents what data is handled, by whom, and for what purpose11,

• Data Protection Impact Assessment (DPIA) evaluates risks in higher-impact processing, such as clinical or genomic data12,

• Transfer Impact Assessment (TIA) assesses legal and technical safeguards for cross-border data13.

For solution providers, familiarity and preparation with these isn’t bureaucracy; it’s evidence of comprehensive risk management and sufficient understanding of complex international compliance. It shortens procurement cycles, and removes one of the most common blockers to large-scale deployments. Teams that can present these materials quickly are viewed as reliable partners who understand both the regulatory and operational realities of handling sensitive patient-derived data.

Latch integrates these same requirements directly into its onboarding and platform architecture. SCCs, DPAs, and supporting assessments are maintained and reviewed with independent auditors, creating a security and documentation layer that travels with your work. This foundation allows solution providers to engage with regulated customers without delay, building credibility while keeping the focus on scientific delivery rather than administrative process.

In Conclusion: Global Data, AI Models, and the New Relevance of GDPR

As science globalizes, GDPR’s reach extends naturally with it. Sequencing data, pathology images, and cellular atlases now cross continents daily to train AI models or power multi-omics pipelines. These datasets almost always contain information traceable to European individuals (directly or indirectly) which places them squarely under GDPR’s jurisdiction.

For solution providers, this means that data sovereignty and privacy controls are now core to scientific scalability. When you train an AI model on European datasets, deploy compute in the U.S., or serve results to a collaborator in Asia, GDPR follows the data at every step.

Far from being an obstacle, this consistency enables collaboration. Standardized privacy frameworks like GDPR create a shared security language between institutions, reducing friction in multinational research.14 They let scientists exchange data and model outputs with confidence that each participant is adhering to compatible privacy and accountability standards.

In this sense, GDPR has become part of the infrastructure of global bioinformatics: a framework that allows sensitive patient-derived data to move across borders, feed algorithms, and return as clinically meaningful insight, all while preserving the trust that underpins scientific progress.

If you are a solution provider interested in white-labeling these tools and offering them to customers through a branded portal, email contact@latch.bio.

LatchBio’s Compliance Team is available at compliance@latch.bio.

• Overview

• Deploying Forch

  • Task IPv6

  • Task Firewalls

  • Node Mounts

  • Task Configuration

  • Task Secrets

  • Node Autoscaling

  • Public IPs

  • Storage Devices

  • Cuda Devices

  • Billing

• Architecture Overview

• Using Forch

Overview

Forch has native support for cross-region (and even cross-cloud) deployments. This brings improved network latency for Latch Pods/Plots and other interactive software for users outside the US west coast, where our compute is currently hosted. It also gives us access to a much larger pool of instances - when the AWS datacenter in Oregon runs out of A100 GPUs, we can opt to get some from GCP in Utah. We can even choose the location by comparing prices of all suitable instances across all regions of all providers, bringing a potentially significant reduction in cost.

It should also be possible to eventually support on-prem deployments as it doesn't make a difference to the orchestrator where the computer is located. Forch is designed to avoid up-front requirements for its compute nodes and their environment. For example, if you never run a task which wants to accept internet connections, you never have to worry about the public IP allocation system or firewall configuration. The Forch components responsible for these features will never run. If the on-prem nodes happen to already have storage devices attached to them, Forch tasks can use the devices without a storage device driver in the orchestrator itself. This means we can automatically support environments without cloud APIs.

As opposed to our Kubernetes setup, Forch does not choose from across pre-defined "node groups". Instead it queries the cloud providers directly for all available instance types and chooses according to a scoring formula. This means that not only is the cost reduced by choosing the best location, even the type of the instance is chosen based on price. Moreoever, Forch uses a generic constraint solver to decide which tasks will run on which nodes, allowing for for very complicated requirements and preferences. For example, you could encode data egress costs (based on estimated task input size) and include them in the price estimate for a given instance. In our Kubernetes set up (if it supported multiple clouds) you would have to manually check the instance prices, estimate egress costs, and make a decision ahead of time. By the time the task was running, the instance prices or availability might also be different, especially when using preemtible spot instances.

We also designed Forch to be easy to debug and maintain. Our experience running Kubernetes has been plagued by errors that are specific to the interaction between various semi-compatible versions of its plugins and the idiosyncracies of AWS EKS (our provider), the Kubernetes resource abstraction, and Go (its implementation language). Commonly, the only error exposed by Kubernetes APIs is the dreaded "context deadline exceeded" which means that something somewhere timed out and we have to scan through millions of lines of logs across dozens of services to find relevant information. Forch is built around a central database which stores the entire history of state transitions for all objects, which are themselves immutable. It is always possible to figure out how the state of the system evolved over time to get to an error. We also include OpenTelemetry-based distributed tracing in all our services to make sure we have very complete, structured information about the actions that Forch takes at the code level.

Deploying Forch

We are working on making Forch as easy to deploy in a third-party cloud account as possible. The current setup involves provisioning an AWS CloudFormation template which creates a setup role, which is passed to a Pulumi deployment to configure a dedicated VPC and other downstream roles. After the initial setup, the CloudFormation role can be optionally revoked as the Forch will use fine-grained roles managed by Pulumi. We expect to provide more flexible deployment options (using an existing VPC, or without any custom AWS IAM permissions) in the future.

In its current form, Forch assumes an IAM role in the target AWS account to issue API calls. We plan to eventually introduce a security barrier in the form of an "Agent" service which runs in the target account itself and acts on behalf of the orchestrator. This service will be fully controlled by the account owner and Forch will no longer require users to grant any permissions to the Latch AWS account.

We currently expect users to use a Latch-provided AWS EC2 AMI for the compute nodes though in practice there are only two requirements for a computer to join the cluster:

1. a Python runtime (typically with uv),

2. an internet connection of some sort (egress only being acceptable).

All features that require cloud APIs are designed to be optional as they may not be necessary for some users and require additional setup, account permissions, and potentially impose limitations on the compute environment. Forch currently supports the following:

Task IPv6

Forch will configure IPv6 addresses for each task if the node's network card reports an assigned IPv6 prefix. This is used by Latch Pods.

Task Firewalls

Forch configures firewalls using nftables on each node to prevent communication between the tasks, the tasks and local networks, and the tasks and the internet. These are fully configurable and allow rules based on specific IPs and task IDs (which resolve to the tasks' current IPv6 addresses). This is used by Latch Pods.

Example configuration (database rows):

task_id=1 peer_task_id=2 direction='ingress'

-- allow task #1 to receive traffic from task #2

Node Mounts

Forch allows tasks to request directories from the node are bind-mounted into the task. This is currently only available to "privileged" tasks. You may request privileged tasks only on nodes for which you have additional authorization explicitly for this feature. Forch currently uses this internally to implement NFS server tasks.

Example configuration (database rows):

-- note: the task must have privileged=true

task_id=1 src_path='/dev' mount_path='/dev' mount_propagation='rshared'

-- allow the task to access the parent node's /dev directory

Task Configuration

Forch allows passing configuration data to tasks in environment variables or injected files. The data is stored in plain text in the Latch-owned central database so is not suitable for secrets or sensitive information. This is used for all Latch workloads.

Example configuration (database rows):

-- task_datum_spec:

id=1 dest_environment_variable='TEST' src_inline='hello world'

id=2 dest_path='/config.json' src_inline='{"hello": "world"}'

-- task_data:

task_id=1 datum_spec_id=1

-- task #1 receives data from spec #1 (TEST='hello world')

task_id=1 datum_spec_id=2

-- task #1 receives data from spec #2 (file /config.json)

Task Secrets

Forch can use external secrets management APIs to securely pass secrets to the tasks in environment variables or injected files. The secrets are absolutely never read outside the compute node running the task requiring the secret. This is used for all Latch Pods.

We provide an AWS SecretsManager driver. This feature requires respective AWS permissions.

Example configuration (database rows):

-- secret:

id=1 provider_id='secret-key' infra_provider='aws' infra_account='000000000000'

-- task_datum_spec:

id=1 dest_environment_variable='KEY' src_secret_id=1

-- task_data:

task_id=1 datum_spec_id=1

-- task #1 receives data from spec #1 (KEY=<REDACTED>)

Node Autoscaling

Forch is able to determine when new nodes are required to run tasks and automatically request new nodes. It is also able to detect empty nodes and request their shutdown after a cool-down period. This is used for all Latch workloads.

We provide an AWS EC2 driver. This feature requires respective AWS permissions.

Forch also allows specifying a schedule anticipating expected demand by day+hour and Forch will scale up nodes in advance to ensure demand can be fulfilled.

Public IPs

Forch can assign public IPs to tasks, if they require routing for incoming connections. We use this internally to implement Latch Pods.

We support nodes with public IPv6 and IPv4 addresses (any number of assigned addresses would work).

We also provide an AWS Elastic IP driver. Forch also provides a custom NAT server to support using Elastic IPs with nodes from a private VPC subnet (i.e. the nodes do not have a pre-assigned public IP). This feature requires respective AWS permissions.

Example configuration (database rows):

-- ip:

id=1 provider_id='eipalloc-00000000000000000' infra_provider='aws' infra_account='000000000000' infra_geo_partition='us-west-2'

-- task_ip:

task_id=1 ip_id=1

-- task #1 will have IP address #1 (provided by AWS Elastic IP with allocation `eipalloc-00000000000000000`)

Storage Devices

Forch allows tasks to mount external devices for persistent storage. This is used in Latch Pods. For ephemeral storage you can use the node's storage devices. This is used for Latch Pods.

We do not currently support placing ephemeral storage on dedicated volumes, but this may change in the future.

We provide an AWS EBS driver which additionally supports volume resizing. We plan to eventually provide volume pre-warming and cross-region/AZ copying. You must create the volumes externally and provide the volume ID to Forch, but this is typically a single API request. This feature requires respective AWS permissions.

Example configuration (database rows):

-- storage_device:

id=1 provider_id='vol-00000000000000000' infra_provider='aws' infra_account='000000000000' infra_geo_partition='us-west-2'

-- task_storage_device:

task_id=1 dev_id=1 mount_path='/mnt' mount_type='filesystem'

-- task #1 will have device #1 (EBS volume `vol-00000000000000000`) mounted to /mnt

-- the node will ensure a filesystem is present on the device before the task starts

Cuda Devices

Forch nodes can attach CUDA devices to tasks, when available. They will also mount the node's CUDA drivers/SDKs into the tasks since the driver version has to match exactly between a container and the host.

Example configuration (database rows):

-- task:

display_name='test' ... dedicated_gpu_type='A10G' dedicated_gpu_count=1

-- task "test" will only run on nodes where an A10G is available

-- the device will be provided to the container along with the CUDA driver

-- using bind mounds from the host

Billing

Forch allows tracking the runtime of tasks and "charging" it to an entity by ID. The task creator must provide authorization to charge to a specific entity. This is used internally by Latch to implement billing.

Architecture Overview

The core components of Forch are:

1. The database (Postgres)

2. The orchestration service

3. The node daemons

4. The NAT servers (per VPC/networking environment)

The orchestrator handles scheduling using constraint solving and executes various environment control API calls e.g. attaching/detaching AWS EBS devices.

In the future, we plan to add an additional "Agent" service which will handle cloud API requests and other environment configuration functions which are currently assigned to the orchestration service. This is to improve permission separation by allowing Forch to function without holding permissions delegated from the compute environment e.g. with an externally-assumable AWS IAM role.

The node daemons send requests to the local container runtime based on tasks read from the database. They also handle system-level setup like firewall configuration (using nftables), storage device preparation (partitioning, file system creation for new devices).

All the services communicate exclusively through polling the database for new rows. Authorization is implemented in-database using row-level security functions, predicated on ED25519 JWKs verified by a plv8-packaged JS implementation. There is no API server, though Latch uses its existing API servers for creating workloads and workload resources (i.e. operating the Forch cluster). User-owned deployments will not need to interact with these services.

The NAT server works by setting up IP-IP tunnels to each node and configuring nftables. This is generally optional, depending on the networking setup. It was originally introduced because the AWS Elastic NAT service cannot be used with Elastic IPs since none of the available Route Table configurations route traffic appropriately. It should be rarely necessary outside of this setup.

Our database schema is based around immutable objects holding configuration for various resources (like provider-specific IDs e.g. AWS EBS volume IDs), and the associated event tables, which hold a parent object ID, timestamp, and type + their event data tables, which hold the data specific to each event type. The state of each object (e.g. if a storage device is attached to a node) is determined by taking the latest event and looking at its type and/or data. We hold the entire history of all state transitions of all objects for observability/debugging purposes.

All services are also traced using OpenTelemetry/Datadog. This is a huge improvement over the infamous Kubernetes/Go "context deadline exceeded" set of errors.

Using Forch

Latch users can look forward to Forch being enabled globally in 2026. For the time being, Forch is already available with explicit opt-in. Users can choose one of several Latch-owned deployments in various AWS regions and we will bring up more regions and other cloud providers over time based on demand. We are working with select partners to make user-owned cloud compute environments generally available by the time Forch rolls out to the public.

If you are interested in Forch, please reach out to contact@latch.bio or on social media.

We may publish detailed posts focused on the engineering behind Forch in the future. Please reach out to kenny@latch.bio or on social media if you are interested so we can prioritize topics accordingly.

Yet understanding the complete state of a cell requires more than measuring RNA. Gene expression is the outcome of a complex regulatory program, influenced by chromatin accessibility, histone modifications, and DNA methylation. Capturing these modalities in their native spatial context is essential for building a more complete atlas of biology.

Go to Data Portal

Today, we are releasing a new data portal with AtlasXomics dedicated to spatial epigenomics. The portal includes datasets from four publications, spanning diseases such as Alzheimer’s and gastric adenocarcinoma, along with studies of heart and brain tissue development.

These resources bring a new layer of information into focus, revealing not only which genes are expressed, but also how the regulatory architecture of the genome shapes that expression in space.

Visit the portal and study these datasets here: https://atlasxomics.latch.bio/data-portal

How does it work?

At the heart of these datasets is a technique called Deterministic Barcoding in Tissue sequencing (DBiT-seq), pioneered by AtlasXOmics. DBiT-seq uses microfluidic chips to flow barcoded oligonucleotides across tissue sections. Each intersection of these flows marks a defined spatial coordinate, which can then be linked to sequencing reads.

The result is a generalizable platform that can add spatial context to a variety of assays. When combined with ATAC-seq, for example, DBiT-seq produces spatially resolved maps of chromatin accessibility. Similarly, paired with CUT&Tag, it reveals the distribution of histone modifications such as H3K27ac or H3K4me3 across tissue.

What does the data tell us?

To understand what insights are revealed through DBiT-seq, let’s walk through an example dataset of how spatial ATAC-seq helps us better understand tumor heterogeneity and chromatin dysregulation in two stomach carcinoma patient samples.

Tumor heterogeneity remains a formidable challenge in cancer research and treatment, manifesting from inter-patient variability to intra-tumor clonal diversity. This heterogeneity complicates understanding of disease progression and therapeutic response. At the molecular level, chromatin dysregulation is a key driver of heterogeneity: aberrant chromatin states influence gene expression, cell identity, and tumor microenvironment (TME) interactions, all of which contribute to resistance pathways and metastasis.

Spatial ATAC-seq combines chromatin accessibility profiling with spatial resolution, allowing researchers to identify regulatory elements active in specific regions of a tissue. By coupling high-throughput sequencing with spatial barcoding approaches, the AtlasXomics platform enables:

• 10µm resolution: Identification of epigenetic states at the single-cell level.

• Spatial context: Mapping chromatin landscapes to morphological features.

• Comprehensive analysis: Integration of gene activity scores, motif enrichment, and peak accessibility with spatial representation.

These capabilities distinguish spatial ATAC-seq from bulk or single-cell ATAC-seq by preserving the physical location of cells, enabling deeper insights into tumor biology.

Mapping cells in TME

This dataset demonstrates how spatial ATAC-seq can disentangle cell state differences between tumor and surrounding normal tissue. By quantifying chromatin accessibility and inferring gene activity scores, it becomes possible to delineate distinct biological compartments and cell types, such as immune, muscle, and tumor-specific populations.

The UMAP projection below highlights major clusters including tumor, mucosa, muscle, and mixed tumor-immune states. When integrated with spatial mapping, these clusters align with morphological regions of the tissue, revealing how chromatin accessibility defines functional neighborhoods within the tumor microenvironment.

Such spatial alignment is essential for understanding how local chromatin states shape cellular interactions, influence immune infiltration, and drive tumor heterogeneity.

Deciphering Spatial Chromatin Dysregulation in TME

Chromatin accessibility reflects the interactions of DNA and transcription factors, and therefore a lens through which one can study underlying cell states in a tumor that lead to specific gene expression programs.

Increased accessibility may allow cancer cells to activate survival programs, upregulating anti-apoptotic genes (e.g., BCL2), stress response regulators, and genes promoting immune suppression (e.g., PD-L1, IDO1/IDO2 ) , enabling adaptation to therapy and ultimately, resistance. Resistant clones can exhibit unique chromatin landscapes with enriched accessibility at loci controlling drug efflux, transcription factors (e.g., MYC), and epigenetic modifiers.

Targeting chromatin regulators, such as histone deacetylases (HDACs) or bromodomain and extraterminal domain (BET) proteins, holds promise as a therapeutic strategy to counter these adaptations and re-sensitize tumors to treatment.

Differential Gene Accessibility Defines Distinct Tumor Compartments

Differential gene accessibility can reveal how distinct tumor compartments are functionally organized. By mapping the spatial distribution of highly differentially accessible genes, researchers can identify how cellular neighborhoods are poised to take on specialized roles of specific TME niches,such as immune hot vs. cold, invasive, or hypoxic environments.

Looking for Drivers of Disease

Examining motif enrichment in tumors compared to matching normal tissue reveals the regulatory elements driving cancer progression. Differential motif analysis reveals transcription factors and expression programs uniquely activated in tumors, shedding light on the epigenetic reprogramming that underpins oncogenesis. This approach can identify key cancer drivers, such as transcription factors supporting unchecked cell growth or immune evasion, and distinguish regulatory pathways associated with therapy resistance.

Additionally, comparing motif enrichment between tumor and normal tissues helps delineate how chromatin accessibility shifts contribute to tumor heterogeneity, highlighting the diverse adaptations of cancer cells to their microenvironment.

By pinpointing these unique regulatory elements, researchers can uncover novel biomarkers for early detection and identify region-specific therapeutic vulnerabilities.

Explore yourself

Spatial epigenomics presents a powerful framework for linking chromatin states to the physical architecture of tissues. By layering accessibility, histone modifications, and transcriptional readouts onto spatial coordinates, these datasets reveal how regulatory programs shape disease progression, tissue development, and therapeutic response.

• Visit atlasxomics.latch.bio/data-portal to visualize and explore yourself.

• Contact colinn@atlasxomics.com to learn how you can perform single-cell spatial CUT&TAG, or ATAC-seq in your experiments.

• If you are a kit or solution provider interested in setting up your own portal and bundle it with analysis packages, contact alfredo@latch.bio

• Release of a 25 million cell spatial atlas spanning all 11 major spatial vendors: including 10X Genomics Visium, Visium HD, Xenium, Bruker Spatial CosMx, STOmics Stereo-seq, Vizgen MERSCOPE, Takara Bio Seeker, AtlasXOmics DBiT-seq, Spatial Genomics GenePS (seqFISH), Element Biosciences AVITI24, and Singular Genomics G4X.

• A public data portal with annotated H5ADs ready to visualize and download

• A spatial vignette to allow researchers to continue with downstream steps, such as cell type annotation, ligand-receptor analyses, or multi-sample differential gene expression

• Tiled AnnData file format and visualization ecosystem to view large counts directly from a blob store without spinning up a server

Engineering biology is moving into an era of data driven and unbiased discovery. Large volumes of high quality measurements are used to tease out targets, disease mechanisms and molecular designs otherwise difficult to identify with the unaided eye.

Spatial assays provide rich information - precise molecular position and paired high resolution images - useful in modeling cell-cell interactions or emergent properties of entire tissues. These additional measurements will be necessary for AI labs to model phenomena beyond binding/structure or for data driven translational teams to mine new hypothesis space.

Scattered across the Internet are thousands of public spatial transcriptomics studies that research everything from 3D reconstruction of human embryos to tumor microenvironments for major cancer types. These datasets might help us develop a new class of therapeutics, but their potential remains locked away due to the expensive human labor required to clean and standardize them.

While there have been previous initiatives for large-scale curation, their efforts have been primarily focused on single-cell sequencing. This leaves statistical models blind to a critical dimension: spatial organization. By linking expression to physical location, we give models “eyes” to learn how cells interact with neighbors, how niches are structured, and how disease architectures emerge. This spatial context transforms fragmented molecular profiles into coherent maps of human biology, potentially opening avenues to therapeutic targets that dissociated single-cell data cannot reveal.

25M Spatial Cell Atlas for Human Anatomy

To tackle the curation challenge, the LatchBio engineering team aggregated 25M spatial transcriptomes from public human studies covering 45 tissue types, 63 diseases and 11 spatial technologies. To our knowledge, this is the largest open access sequencing based spatial atlas available to date.

The team used latch-curate, an human-in-the-loop curation framework that boosts annotation quality and consistency by incorporating information from entire papers and unstructured supplements.

A detailed whitepaper on latch-curate's design and capabilities is available at latch.bio/latch-curate.

Each dataset is published on console.latch.bio/data-portal as an annotated H5AD, structured with standardized ontologies for assay, cell type, condition, disease, tissue, patient ID (if available), and organism. Author annotations from the original paper are also included when available.

Spatial toolkit for scientists to go further

While curated datasets provide the substrate for scientific inquiry, more sophisticated downstream analyses are often required to extract knowledge and generate new hypotheses.

The Latch spatial toolkit provides scientists with step-by-step workflows to further analyze curated datasets or new studies from their own raw data.

The analysis can be broken down into 5 steps:

1. Counts construction from raw instrument data

2. Image alignment

3. Quality control, filtering, transformation

4. Multi-sample differential gene expression with pseudo-bulk analyses

5. Cell type annotation

6. Ligand-receptor analysis

Counts construction

Through partnerships with solution providers such as Takara Bio and AtlasXOmics, we build pipelines that convert raw instrument data into Scanpy-ready H5AD files. Alongside these, we generate PMtiles of morphology, H&E or IF stains, and transcript images. This collection can be loaded directly into the H5 viewer in Latch Plots for interactive exploration.

Image alignment

Spatial analysis requires precise segmentation of tissue regions. Once an H5AD is opened in the viewer, scientists can select images, define anchor points using the landmark system, and apply alignment. Both affine and STAlign modes are supported, with the latter designed for samples with severe tissue distortion.

Quality control, filtering, and transformation

Within the Plot notebook, scientists can run multi-sample QC, filter cells, detect highly variable and spatially variable genes, and apply Leiden clustering to define spatial domains of interest.

Multi-sample differential gene expression with pseudo-bulk analysis

After QC, scientists can launch pipelines such as cell2location for deconvolution in spot-based assays, RTCD, or marker-based approaches referencing PangloDB. These methods enable flexible annotation across technologies.

Ligand-receptor analysis

Once cell types are annotated, the next question is how they interact. Squidpy’s ligand-receptor module evaluates large numbers of spatial permutations, often at high computational cost. To increase focus and efficiency, scientists can use the viewer’s lasso tool to define regions of interest, ensuring interaction analysis captures only spatially proximal, biologically plausible events. Results are returned as plots of ligand–receptor pairs between cell types.

anntiles: tiled AnnData served directly from blob stores

To see a bird’s-eye view snapshot of the results generated by these tools, navigate to our Data Portal (console.latch.bio/data-portal) and click “Visualize” on our top featured datasets.

Our team has created a tiled AnnData file format (`anntiles`) and a suite of visualization tools that enable exploration of very large matrix-based counts data without spinning up a server. These tools will be released soon and are used to implement the notebooks exposed through the link above.

If you are a scientist interested in accessing these toolkits, please reach out and we will send you a link to access!

If you are a solution provider interested in white-labeling these tools and offering them to customers through a branded portal, contact alfredo@latch.bio. If you are interested in co-developing exciting new spatial tools and collaborating on our roadmap reach out to hannah@latch.bio.

What is a latch pod?

Latch pods are scientific computing environments built for biology. Pods let you run biological data science and machine learning applications in one place.

• A central system for JupyterLab, RStudio, and VSCode environments.

• A common set of tools for intuitive sharing of R and Python code.

• Collaborative notebooks for single-cell, spatial, image viewing, and more.

• Flexible control over CPUs, memory, and GPUs for different projects and users.

• Cloud billing features to monitor spending and set daily limits.

What is bring-your-own-compute?

Bring-your-own-compute is a new Latch architecture that allows you to:

• Connect your AWS environment

• Leverage your existing compute

• Specify geographic regions

• Use your cloud credits

All without data or compute leaving your servers.

How does it work?

High-level architecture

Bring-your-own-compute is part of a new latch architecture, which operates with a control plane and compute plane.

• The control plane includes services that Latch manages in your account.
  (e.g. Latch pods.)

• The compute plane is where your data is processed. There are two types of compute:

  • Latch Managed compute, which runs in your Latch account.

  • Bring-your-own compute, which runs in your AWS account.

Each Latch workspace has a storage bucket, which is in your AWS account.

Motivation and roadmap

We built this to deliver several benefits to our customers:

1. Ensure that data never leaves your cloud tenancy.

2. Allow you to use existing cloud provider credits.

3. Reduce network latency for users geographically far from us-west-2

4. Increase GPU availability by allowing compute hosts from other AWS regions

5. Improve cost efficiency by exposing a more diverse set of compute options

Based on interest, we’ll plan to continue supporting bring-your-own compute for the entire platform, including workflows, nextflow, and plots. Long-term, we’ll move towards support for GCP, Azure, and more compute providers.

How do I get started?

To get started, please comment or send a note to kyle@latch.bio and we’ll send you instructions. Thanks for reading.

This session features a live walkthrough of Plots—an early-access, browser-based tool for interactive spatial and single-cell analysis. Whether you're already using Seeker™ kits or exploring spatial biology for the first time, you'll get a practical look at how to unlock biological insights from complex datasets.

What you’ll learn:

• Remove off-tissue beads using background filtering for Seeker™ data

• Align and overlay H&E or immunofluorescence images onto spatial embeddings

• Explore and compare H5AD files across multiple samples

• Identify spatially variable genes and regions

• Detect differentially expressed genes across tissue zones

• Run ligand-receptor analysis in spatially defined domains

• Perform cell type annotation and deconvolution

• Map cellular trajectories in tissue context

Register now to see how these analyses work in practice.

Register for Webinar

Register for Webinar

What to Expect

The webinar will cover methods for:

1. Image Alignment for Seeker™ datasets with STAlign.

   Align serial-section H&E images to the spatial representation of the tissue.

2. Differential Gene Expression Across Spatial Regions

   Identify differentially expressed genes across Seeker™ or Trekker™ tiles.

3. Ligand-Receptor Analyses

   Detect communication between neighboring cell types using curated ligand-receptor pairs within selected regions.

4. Cell Type Annotation and Deconvolution

   Resolve cell-type composition from mixed transcriptomes using spatial-aware reference models.

…and more

Register to see a live demo of these analyses.

Register for Webinar

Register for Webinar

What You’ll Learn

Plots enables scientists to analyze complex spatial omics data through an intuitive, no-code interface. In this session, you’ll learn how to:

• Perform background removal

• Perform QC, filtering, and Leiden clustering

• Overlay H&E or immunofluorescence images on spatial embeddings

• Visualize and interact with H5AD files across multiple samples

• Use Lasso selection to define regions of interest

• Identify spatially variable genes

• Detect differentially expressed genes

• Conduct ligand-receptor analysis within selected spatial domains

• Perform cell type annotation and deconvolution

• Explore cellular trajectories in spatial context

Everything is accessible through a user-friendly graphical user interface. No programming required.

We look forward to seeing you there!

The past decade has witnessed many advancements in the scale and type of modality we can measure. Take the example of Takara Trekker, a clever technique to tag each nucleus within its native tissue environment with unique spatial barcodes using UV light. Or Parse Biosciences’ Penta kit, which uses a split-and-pool approach to barcode one to five million cells without the need to use microfluidics instruments for physical single-cell isolation. Even large-scale proteomics workflows, such as those from Olink or CITE-seq, now rely on antibodies conjugated to unique DNA sequences to convert protein abundance into a sequencing-compatible readout.

Whether you're measuring gene expression, chromatin state, protein abundance via DNA tags, or spatial context, the final step in these workflows is almost always the same: sequencing.

Originally built to read the human genome, modern sequencers have become general-purpose molecular readout devices—a universal endpoint for biology.

As of June 2025, the market features 37 sequencing instruments. To help navigate this landscape, we’ve mapped out the 10 key companies building these platforms, compared them across core technical attributes, and published a database to make the data easier to explore and debate.

You can access the database here, and if you see an error or want to contribute, leave a comment. This is meant to be a living and collaborative resource for the community.

In this article, we will explore:

• From Sanger to Today: The History of Sequencing Technologies

• Key Players in 2025: A Quick Rundown of Their Latest Chemistries

  • Oxford Nanopore: ONT’s Q30+ Kit14 (launched 2024)

  • Pacific Biosciences: HiFi Chemistry on Revio (launched 2023)

  • Pacific Biosciences: SPRQ Chemistry (launched late 2024)

  • Element Biosciences: Cloudbreak UltraQ (launched 2024)

  • Illumina: XBS-LEAP (launched 2023)

  • Complete Genomics: DNBSEQ with CoolMPS and HotMPS Chemistries (mature by 2025)

  • GeneMind: Lightning Chemistry (launched 2024)

  • Ultima Genomics: Solaris with ppmSeq (launched 2025)

• Comparison by Attribute

  • The U.S. seems to be trailing behind Chinese manufacturers in terms of throughput and speed.

  • Scientists need to navigate the cost per genome vs instrument price paradox.

  • Is Cost per GB all that matters? Beware of hidden infrastructure costs.

  • Multi-omics Support: Are sequencing machines evolving into something more?

We hope this will become a comprehensive resource for researchers as they navigate the NGS landscape in 2025.

From Sanger to Today: The History of Sequencing Technologies

1977 - 2005: The Foundation Years

The story of DNA sequencing begins in 1977 with Fred Sanger's chain-termination method, a breakthrough that first made reading the language of life possible. Sanger's initial triumph was modest by today's standards, decoding the 5,000-base genome of bacteriophage φX174, but it represented humanity's first glimpse into the digital code underlying biology. This pioneering method relied on clever chemistry: DNA polymerase would incorporate modified nucleotides that terminated chain extension, creating fragments of different lengths that revealed the sequence through gel electrophoresis.

For nearly three decades, Sanger's approach dominated the field through incremental improvements. The 1980s and 1990s brought automation via capillary sequencers, replacing labor-intensive gel systems with continuous processing machines. Yet even automated systems remained fundamentally limited; they were accurate but painfully slow, requiring years and tens of millions of dollars to sequence a single human genome. The Human Genome Project, completed in 2003, represented this era's culmination: a heroic international effort that demonstrated both genomics' transformative potential and the desperate need for faster, cheaper technologies.

2005 - 2010: The NGS Revolution

The mid-2000s brought the advent of “next-generation sequencing” (NGS), characterized by massively parallel short reads. 454 Life Sciences introduced a pyrosequencing platform in 2005 (read lengths ~100s of bases), and in 2006-2007 Solexa/ Illumina launched its sequencing-by-synthesis (SBS) platform, followed by Applied Biosystems’ SOLiD. These second-generation sequencers could generate gigabases of data in days, massively reducing the cost per base (dropping from ~$10,000 per megabase to mere cents). Illumina’s SBS technology quickly came to dominate the market, due to its accuracy and throughput. Over the past decade, Illumina has at times held ~80% of the sequencing market share.

2010s: The Rise of Long Reads (Third-Generation)

In the 2010s, third-generation sequencing technologies emerged, distinguished by the ability to sequence single molecules and produce much longer reads (thousands to tens of thousands of bases).

Pacific Biosciences pioneered this transition with their Single Molecule Real-Time (SMRT) sequencing platform in 2011. Their zero-mode waveguides created optical chambers where individual DNA polymerases could be observed incorporating fluorescent nucleotides. Oxford Nanopore Technologies took a different approach, threading DNA through protein nanopores and measuring electrical signal changes as bases passed through. When ONT's MinION launched mid-decade, it brought sequencing capabilities to a USB-stick-sized device.

These long-read platforms initially faced skepticism due to higher error rates, but the errors were random rather than systematic, making them correctable through coverage or consensus approaches. More importantly, long reads solved problems short reads couldn't address: de novo genome assembly of complex regions, large structural variant detection, and full-length isoform sequencing. As accuracy improved (PacBio's HiFi reads achieving over 99.9% accuracy, ONT's newer chemistries reaching ~99% single-read accuracy), long reads transitioned from specialized tools to mainstream platforms.

By the late 2010s, the cost of sequencing a human genome had fallen below $1,000, a milestone achieved well ahead of the trajectory of Moore’s Law.

2020s: Toward Multi-omic, Spatial, and Ultra-High-Throughput Sequencing.

We are now in a new phase defined by multi-omic compatibility, spatially-resolved sequencing, and ultra-high throughput machines. Sequencing is being integrated with other modalities: for example, Illumina and Singular Genomics are developing workflows to sequence not just genomes, but also spatial transcriptomics. Illumina also recently acquired SomaLogic, signalling its movement into the proteomics field.

Throughput has also exploded. Traditional “high-throughput” sequencers output on the order of a few terabases per run, but new instruments are extending this by an order of magnitude. Illumina’s latest NovaSeq X series, for instance, can output up to 16 terabases of data in a single run (26 billion reads per flow cell), and emerging players are pushing costs down further (more on this below). There is also a convergence of technologies: short-read companies are adding long-read or synthetic-long-read capabilities, long-read companies have launched short-read platforms, and entirely new chemistries are being introduced. All of this means researchers in mid-2025 face a rich but complex menu of sequencing options.

Key Players in 2025: A Quick Rundown of Their Latest Chemistries

Oxford Nanopore: ONT’s Q30 Duplex Kit14 (launched 2024)

Oxford Nanopore Technologies (ONT) sequences DNA using biological nanopores: tiny protein channels embedded in a membrane that detect DNA molecules as they pass through. As a single-stranded DNA molecule moves through the pore, it causes disruptions in ionic current. These changes are measured and interpreted by machine learning models to identify the underlying nucleotide sequence in real time.

Traditionally, ONT relied on simplex sequencing, where only one strand of a DNA molecule is read. While this allows for extremely long reads, often tens of kilobases or more, simplex reads suffer from several challenges. The signal measured by the pore reflects a short k-mer (typically 5 or 6 bases), making it difficult to resolve individual bases when homopolymers or repetitive regions are present. Because only one strand is read, there is no independent validation of each base, so errors-especially insertions and deletions-tend to persist.

As a result, simplex reads historically achieved modal accuracies of only Q15 to Q18, or about 97 to 98 percent accuracy.

The introduction of Q20+ Kit14 brought ONT’s first broadly available duplex sequencing chemistry. In duplex mode, both strands of a double-stranded DNA molecule are sequenced in succession using a specially designed hairpin adapter. The basecaller then aligns the two reads and compares them to correct random errors and resolve ambiguous regions. Because the errors in each strand are largely uncorrelated, this reconciliation process significantly improves confidence in each base. Simplex reads now achieve around Q20 (~99%) accuracy, while duplex reads regularly exceed Q30 (>99.9%), rivaling the performance of short-read platforms in base-level precision.

Practically speaking, this means that ONT with Q20+ Kit14 can now be used for workflows previously out of reach for nanopore due to accuracy concerns, such as low-frequency variant detection, cfDNA sequencing, or methylation-aware diagnostics, while still enabling applications Illumina cannot touch, like ultra-long read assemblies, haplotype phasing, or native DNA/RNA modification analysis.

Pacific Biosciences: HiFi Chemistry on Revio (launched 2023)

Pacific Biosciences (PacBio) developed a sequencing method called single-molecule real-time (SMRT) sequencing, which watches DNA being copied in real time, one molecule at a time. In this system, a single DNA polymerase enzyme is anchored inside a tiny well called a zero-mode waveguide (ZMW). These wells are designed so that only the very bottom, where the enzyme is working, is illuminated. As the polymerase adds each nucleotide, which carries a unique fluorescent tag, the system detects a quick flash of light and records which base was incorporated.

This technology allows PacBio to read very long stretches of DNA without needing to break them up into short fragments, but early versions of SMRT sequencing had a major drawback of low accuracy.

Because each molecule was only read once, errors from things like fluorescence misreads, dye noise, or enzyme slippage made the data noisy. Single-pass reads often had error rates around 10-15%, especially in homopolymer regions or sequences with complex structure.

To solve the accuracy problem in its original SMRT sequencing, PacBio developed HiFi reads (short for High-Fidelity reads), which combine the length advantages of long-read sequencing with the accuracy of short-read platforms.

Instead of reading a DNA molecule once, PacBio circularizes each DNA fragment using special adapters, forming a loop called a SMRTbell template. The DNA polymerase then continuously reads around the circular molecule, passing over the same insert multiple times, usually 10 to 20 passes. Each pass produces a slightly different version of the sequence due to random errors. By averaging these multiple observations, PacBio’s algorithm generates a circular consensus sequence (CCS) that is highly accurate.

This technique drastically improves read quality. HiFi reads typically range from 10 to 25 kilobases in length, and because they’re built from multiple subreads, they reach Q30 to Q40 accuracy (99.9-99.99%).

Starting at a modest $600M in revenue in 2023, long-read sequencing is projected to reach $1.34B in 2026, likely due to increased market trust and accuracy.

Pacific Biosciences: SPRQ Chemistry (launched late 2024)

Building on the success of HiFi sequencing, PacBio took a bet on their first multi-omics chemistry: SPRQ (pronounced “spark”), designed to extract both DNA sequence and regulatory information from the same molecule.

Unlike bisulfite or enzymatic methods that chemically modify or fragment DNA, SPRQ preserves long, native DNA molecules. A transposase enzyme is used to preferentially insert special adapters into open (accessible) regions of the genome. These adapters include 6-methyladenine (6mA) marks. Open chromatin tends to be more accessible to the enzyme, so this step selectively “labels” active regulatory regions.

As with standard HiFi sequencing, these DNA fragments are circularized into SMRTbell templates, which allows the polymerase to read the same molecule multiple times for high-fidelity consensus sequencing. PacBio’s polymerase begins copying the circular DNA in real time.

As it passes through modified regions (like 6mA or naturally occurring 5mC), the speed of base incorporation changes slightly. These kinetic "pauses" or slowdowns are subtle but measurable. The sequencing software detects these polymerase slowdowns and matches them to specific base positions. It then combines: base sequence from HiFi reads (Q30-Q40), 6mA tagging patterns (open chromatin), and methylation signals like 5mC (native epigenetics). In just one sequencing run, SPQR provides high-accuracy DNA sequence, maps of chromatin accessibility, and detection of base modifications like 5mC and 6mA.

Element Biosciences: Cloudbreak UltraQ (launched 2024)

Element’s AVITI system uses sequencing-by-synthesis (SBS), but with a unique chemistry built around rolling circle amplification (RCA). Instead of reading a single DNA strand, AVITI circularizes DNA fragments and amplifies them into dense clusters called polonies, thousands of identical copies of the same DNA molecule in one spot. As the sequencer adds fluorescently labeled bases one at a time, it captures images and averages the signal across all the identical copies in the cluster. This built-in redundancy helps reduce random noise and improves overall base-calling accuracy.

In 2024, Element introduced Cloudbreak UltraQ, a high-accuracy upgrade to its standard chemistry. UltraQ builds on the same core principles but increases the number of observations per base and improves the enzymes and dyes used during synthesis. This results in sharper signals and cleaner base incorporation, which the AVITI’s software can use for even more precise error correction. With UltraQ, the system reaches Q50+ accuracy (just one error in 100,000 bases!)

Illumina: XBS-LEAP (launched 2023)

Illumina’s newest sequencing chemistry (as of 2023-2025) is XLEAP-SBS™, which was

introduced with the NovaSeq X series instruments. XLEAP-SBS (previously code-named “Chemistry X”) is an overhauled SBS chemistry engineered for faster cycle times and higher accuracy. Not a lot of technical details are available online on how the chemistry actually works, but Illumina claims the following improvements: a new thermostable polymerase, optimized fluorescent terminator nucleotides, and upgraded clustering chemistry on patterned flow cells . The result is roughly 2× faster incorporation speed per cycle and up to 3× greater basecalling accuracy. In fact, NovaSeq X flow cells can generate ≥85% of bases at Q40 or higher (99.99% accuracy), in comparison to the Q30 standard of prior sequencers. By

late 2023 Illumina enabled 2×300 bp reads on NovaSeq X flow cells (primarily for microbiome and metagenomics users needing longer coverage). Additionally, Illumina has back-ported XLEAP-SBS to mid-range instruments. An early 2024 update allows NextSeq 1000/2000 systems to use XLEAP-SBS with new P4 flow cells (~500 Gb output), indicating Illumina’s intent to make the chemistry standard across its lineup.

Complete Genomics: DNBSEQ with CoolMPS and HotMPS Chemistries (mature by 2025)

Complete Genomics’s sequencing platforms are built around a core technology called DNA Nanoball sequencing (DNBSEQ). Unlike other short-read platforms that use bridge amplification, MGI circularizes DNA fragments and amplifies them using rolling circle replication. This produces compact DNA nanoballs, each consisting of hundreds of identical copies of a single DNA molecule. These nanoballs are then loaded onto a patterned flow cell at high density, with each occupying a precise, separate position. Because each nanoball contains many copies of the same fragment, signal averaging becomes highly effective, allowing for accurate base detection with minimal cross-talk between spots.

Traditionally, MGI sequencers used a sequencing-by-synthesis (SBS) approach similar to Illumina’s, with fluorescently labeled nucleotides added one base at a time and imaged to determine the sequence. However, MGI has developed two distinct chemistries to further improve speed, accuracy, and data quality: HotMPS and CoolMPS.

HotMPS is an updated version of SBS that uses directly labeled nucleotides with enhanced polymerase enzymes. The chemistry has been optimized to reduce incorporation bias and fluorescence noise, delivering faster cycle times and improved base-calling accuracy. It's the default chemistry for high-throughput runs on platforms like the T7 and T20, where large-scale projects require both speed and scalability.

CoolMPS, by contrast, takes a different approach. Rather than using fluorescently labeled nucleotides, CoolMPS incorporates unlabeled bases during sequencing. After each incorporation step, fluorescently labeled antibodies that specifically recognize the incorporated base are added to detect which base was added. This indirect detection strategy reduces chemical "scarring" and dye-related artifacts, resulting in cleaner signal and longer read lengths with fewer errors. Because the nucleotides themselves are unmodified, polymerase activity is more consistent and incorporation fidelity is higher.

Together, these chemistries support read lengths up to 2×300 bp, with over 85% of bases typically achieving Q30 quality. On the highest-end system, the DNBSEQ-T20×2, MGI platforms can output up to 72 terabases in a single 80-hour run, currently the most in the industry! This scalability, paired with high accuracy and unique chemistry options, makes MGI a strong contender in applications like population-scale genomics, clinical diagnostics, and spatial transcriptomics.

GeneMind: Lightning Chemistry (launched 2024)

GeneMind’s SURFSeq Q uses classic sequencing-by-synthesis (SBS) on patterned flow cells, where DNA fragments amplify into dense, equally spaced clusters across the chip. Each cycle, fluorescently labeled nucleotides are added to the clusters, imaged, and then cleaved to read the next base, the same core process as Illumina-style sequencers.

To significantly boost both speed and quality, GeneMind introduced Lightning Chemistry in 2024. This chemistry improves three core areas: signal clarity, cycle speed, and base-calling accuracy. Brighter fluorescent labels and refined polymerase enzymes allow faster and cleaner nucleotide incorporations. The combination enables ultra-high throughput. SURFSeq Q delivers up to 14 Tb per run (≈9 Tb/day), while maintaining exceptional accuracy, with 90% of bases at Q40 or higher.

Ultima Genomics: Solaris with ppmSeq (launched 2025)

Ultima Genomics takes a unique approach to short-read sequencing by replacing traditional flow cells with a spinning silicon wafer etched with billions of tiny nanowells. Each well is designed to hold a single DNA fragment, which is anchored using special adapters during library preparation. Once loaded, the wafer spins rapidly to distribute sequencing reagents evenly across its surface, a process that also minimizes fluidic complexity and speeds up reaction cycles.

Ultima’s sequencing chemistry follows a method called sequencing-by-addition. Instead of flowing in all four nucleotides (A, T, C, G) at once (as is common in sequencing-by-synthesis platforms) Ultima flows only one base at a time. For example, in the first cycle, only adenine (A) nucleotides are introduced. If the complementary base on a DNA fragment is thymine (T), the polymerase incorporates an A and emits a fluorescent signal. If not, no incorporation occurs and no signal is recorded. This single-base flow cycle is repeated sequentially (A, then C, then G, then T) across hundreds of cycles to reconstruct the DNA sequence one base at a time.

To enhance basecalling accuracy, especially in error-prone regions like homopolymers, Ultima employs a proprietary method called ppmSeq (plus/minus strand sequencing). This technique sequences both strands of the same DNA molecule independently. The forward and reverse reads are then compared, allowing the system to detect and correct mismatches or insertion-deletion (indel) errors. By cross-validating the two strands, Ultima improves variant calling accuracy and read confidence without sacrificing speed or cost.

Thanks to its simplified chemistry and real-time error correction, Ultima achieves variant-calling performance comparable to short-read leaders like Illumina, with base-level Phred quality scores approaching Q30-Q40, or 99.9% to 99.99% accuracy.

Singular Genomics

Singular Genomics’ G4X platform introduces a hybrid of high-throughput sequencing and spatial biology in a single system. At its core, G4X uses sequencing-by-synthesis (SBS), where fluorescently labeled nucleotides are incorporated one base at a time and recorded by an optical system. But what sets G4X apart is its ability to perform in situ sequencing directly on preserved tissue samples, using a method called Direct-Seq.

With Direct-Seq, FFPE (formalin-fixed paraffin-embedded) tissue sections can be sequenced without the need to extract and convert RNA into cDNA. Instead, the nucleic acids remain physically anchored in the tissue, and sequencing happens right where the molecules naturally reside. This is achieved by flowing specially designed polymerases and fluorescent nucleotides over the sample. As the enzyme incorporates each nucleotide, it emits a flash of light that is recorded by the instrument’s camera, identifying the added base and its precise location within the tissue.

This direct sequencing preserves spatial context while revealing gene expression, allowing researchers to detect RNA transcripts without disrupting the surrounding tissue architecture. In parallel, the G4X system can also perform protein detection using fluorescently tagged antibodies and identify genomic variants, such as single nucleotide changes or small insertions and deletions, from the same section.

By layering RNA, protein, and mutation data into one integrated dataset, G4X is the first to enable full spatial multi-omic profiling in the same sequencer.

Short-read platforms are adding long-read capabilities.

Historically, short-read sequencers like those from Illumina dominated the market with high accuracy and massive throughput, but they struggled to resolve complex regions of the genome due to limited read lengths. To overcome this, Illumina introduced Complete Long Reads (CLR), a synthetic long-read approach that uses molecular barcodes to tag long DNA fragments, which are then sequenced as short 2×150 bp reads and computationally stitched back together.

This enables users of NovaSeq X and 6000 to generate long-read-like data without investing in new hardware. The reconstructed long reads achieve Q30-Q40+ accuracy (99.9-99.99%).

Other short-read companies are following a similar strategy. Element Biosciences, for instance, acquired Loop Genomics to offer barcode-based long-read reconstruction on the AVITI platform. The platform’s native short-read chemistry already reaches Q40 accuracy (99.99%), making it well-suited for high-precision long-read assembly.

These developments reflect an interesting trend. Short-read platforms are expanding into long-read territory by offloading complexity to molecular barcoding and software, while preserving the accuracy and cost-efficiency advantages of short-read sequencing.

Long-read platforms are also doing short-reads.

Oxford Nanopore Technologies (ONT) introduced Short Fragment Mode to support DNA fragments as short as 200-500 bp, enabling workflows with degraded or low-input samples like FFPE tissue or microbial DNA. Combined with Duplex Sequencing, which boosts accuracy by sequencing both strands of a molecule, ONT now supports applications once reserved for short-read platforms.

PacBio launched Onso, a dedicated short-read sequencer using a novel sequencing-by-binding chemistry. Instead of incorporating nucleotides, this approach detects fluorescently labeled bases as they bind reversibly to the template strand, reducing signal noise and improving accuracy. Onso achieves Q40+ accuracy and is aimed at precision-focused markets like oncology and NIPT.

Though these are technically impressive developments, it’s still unclear whether long-read companies can compete head-on in the short-read market. ONT’s and PacBio’s efforts appear more like adaptations to support existing users, those who already own a MinION or Revio, and want to handle occasional short-read workloads, rather than serious attempts to replace Illumina or AVITI for core short-read use cases.

Comparison by Attribute

Disclaimer: This section is built on our CSV, which aggregates public and online information. While some providers have transparent spec sheets on their websites, others don’t and require direct sales contact. In such cases, we tried our best to provide estimates. Please reach out to hannah@latch.bio to propose any corrections.

The U.S. seems to be trailing behind Chinese manufacturers in terms of throughput and speed.

One headline metric is maximum data output (throughput). This determines how many samples or how deep a sequencing run can go in a given time.

The current champions in output are Complete Genomics’s DNBSEQ-T20 (or T10 or T7), Illumina’s NovaSeq X, GeneMind’s SURFSeq Q, ONT’s PromethION 48, and Ultima’s UG 100 - all capable of multi-terabase outputs per run:

• DNBSEQ-T Series: Topping the chart is Complete Genomics, a subsidiary of Complete Genomics in China, which dominates when it comes to throughput per day. Their DNBSEQ-T20 instrument yields 72 Tb data for one 80-hour run, translating to ~21.6 Tb/day. The T10 variant achieves even higher total output at 76.8 Tb per run over 106 hours (~17.4 Tb/day), while the T7 offers 7 Tb in 24 hours (7 Tb/day).

• SURFSeq Q: GeneMind's high-end platform achieves an impressive 14 Tb total output in just 36 hours, translating to over 9.3 Tb/day. This positions it as the second-fastest daily throughput platform, though with lower total capacity than the largest MGI systems. A smaller SURFSeq configuration matches the DNBSEQ-T7 at 7 Tb/day.

• NovaSeq X Series: Illumina's flagship production platform delivers up to 16 Tb per run in 48 hours, achieving 8 Tb/day at peak configuration. Multiple flow cell options provide flexibility, with smaller configurations generating 3-8 Tb per run. The platform balances throughput with established ecosystem support, though it trails MGI's peak daily rates.

• UG 100: Ultima Genomics' cost-focused platform generates 3.6 Tb per run in 12-14 hours, achieving approximately 6.2 Tb/day. While lower in absolute output, this platform targets the $100 genome milestone through exceptional cost efficiency rather than maximum throughput.

• PromethION Series: Oxford Nanopore's long-read flagship achieves 7 Tb total output but over 72-hour runs, resulting in 2.3 Tb/day for the 24-flow cell system. The 2-flow cell variant produces 580 GB over similar timeframes. While lower in raw throughput, these platforms enable applications impossible with short-read systems through ultra-long reads exceeding up to 4 MB.

In the upper quadrant of the market, we have Ultima Genomics’ UG100 ($1.5M), Illumina’s NovaSeq X ($985,000) and NovaSeq X Plus ($1.25M), with NovaSeq 6000 having a similar price tag as NovaSeq X, but with a different range of throughput. Complete Genomics’s DNBSEQ-T7 especially stands out, since it delivers 7 Tb/day—nearly matching NovaSeq X Plus’s 8 Tb—but costs only ~$600K, half the price. Even in the mid-throughput range, DNBSEQ-T1+ comes at $299,000, almost a third of the cost of a NovaSeq 6000. (Note: Complete Genomics doesn’t publish its costs online, but these prices have been confirmed by a representative from Complete Genomics)

The market is heavily clustered in the low-to-mid throughput and cost range. Let’s zoom into them.

In the low-throughput range, most instruments output under 500 Gb/day, with a wide spread in pricing. Illumina offers the broadest lineup—from the low-cost iSeq to the mid-tier NextSeq 1000—ranging from ~$20K to over $300K. Oxford Nanopore and Thermo Fisher offer alternatives like MinION, GridION, and Genexus, with varying form factors and costs around $100K–$300K.

Element Biosciences and Complete Genomics enter at slightly higher throughput levels (100–400 Gb/day) with machines priced near $300K–$400K, offering better output per dollar. PacBio’s Revio is the clear outlier (producing ~400 Gb/day at ~$600K) but includes HiFi accuracy and native methylation calling, justifying its premium.

Scientists need to navigate the cost per genome vs instrument price paradox.

The most expensive machines often deliver the lowest cost per genome, while affordable instruments can carry surprisingly high per-sample costs. You can achieve low cost per gigabase, but only with instrument investments in the millions of dollars range. Looking at cost per genome alone can be misleading.

Below, we compare cost per genome across various providers on the market. To calculate cost per genome, we:

• Gather maximum data yield per run from instrument provider websites

• Obtain maximum reagent cost per run from provider websites or news report estimates when unavailable

• Calculate cost per gigabase = maximum reagent cost ÷ maximum data yield

• Assume 30X genome coverage requiring approximately 100 Gb of sequencing

• Calculate cost per genome = cost per gigabase × 100

In the realm of short-read sequencing, leading this race is Ultima's $1.5M UG 100, which achieves the lowest cost at $80 per genome, while Illumina's $1.25M NovaSeq X delivers $103-268 per genome depending on configuration.

Complete Genomics doesn’t sell their DNBSEQ-T20 directly. Instead, pricing is amortized over reagent contracts (e.g., $0.99/Gb for 72 Tb/run or $99/per genome), good for mega-scale centers sequencing >50k genomes/year.

In the middle tier, platforms like the NovaSeq X 1.5B, DNBSEQ-T7, and AVITI strike a balance between upfront investment and per-genome cost. These instruments typically fall in the $300K to $1M range and bring sequencing costs down to around $150-300 per genome. They’re a good fit for labs that don’t need ultra-high throughput but still want to process thousands of samples a year efficiently. Many offer flexible flow cell configurations, so you’re not locked into massive runs to make the economics work.

Element Biosciences' AVITI emerges as a fierce contender in the mid-range market. At $289,000, the instrument delivers $200 per genome costs that directly compete with Illumina's NovaSeq X 25B ($206 per genome) while requiring less than one-third the capital investment ($985K for NovaSeq X). Even AVITI's lower-throughput configurations maintain competitive economics, with cost per gigabase ranging from $5.60-$7.20, closely matching the NovaSeq 6000's $4.55-$6.23 range across S1, S2, and SP flow cells. This positioning allows institutions to access near-production-scale economics without the substantial upfront investment typically required for high-throughput sequencing.

While short-read platforms continue to drive costs down across a range of throughput tiers, long-read technologies are following close behind, offering increasingly competitive pricing with the benefit of spanning complex regions.

Long-read whole-genome sequencing costs are difficult to pin down because they hinge on several interacting factors. While Oxford Nanopore advertises roughly $345 for about 100 Gb and PacBio quotes around $500 for ~30× human genome coverage on its Revio system, actual price depends on flow-cell yield (Nanopore flow cell output ranges from ~100 Gb to 290 Gb), the accuracy mode chosen (cheaper Q20 simplex reads versus higher-quality Q30+ duplex reads), and application-specific depth or accuracy requirements.

Is Cost per GB all that matters? Beware of hidden infrastructure costs.

Converting electrical signals and optical data into usable format is no easy task. While cost per gigabase has become the dominant metric for comparing sequencing platforms, this narrow focus obscures the substantial computational infrastructure investments required to transform raw sequencing data into useful insights.

Once bases are called, labs must invest in compute infrastructure, software licenses, data transfer pipelines, storage, and ongoing maintenance to extract value from the output.

What do these hidden costs include?

1. Primary analysis: basecalling, demultiplexing

2. Secondary analysis: alignment, variant calling, assembly

3. Data storage: terabytes of raw + processed data per run

4. Compute infrastructure: on-prem clusters or cloud credits

5. Software licenses: proprietary tools (e.g. DRAGEN)

6. Data transfer: ingress/egress fees for cloud platforms

7. Compliance: secure, reproducible pipelines for CLIA/CAP/ISO especially for diagnostics use cases

At AGBT 2025, Illumina CEO Jacob Thaysen captured this shift:

"Customers are not only buying a product anymore…they're buying into a whole ecosystem… applications, workflows, informatics… That is the game."

Below, we compare how major sequencing vendors address these hidden infrastructure costs through engineering design and software strategy.

DRAGEN: FPGA-accelerated genomics

Illumina recognized the challenges in storage and processing of genomic data early, acquiring Edico Genome and the DRAGEN Bio-IT platform in 2018 to architect a solution.

DRAGEN uses highly reconfigurable field-programmable gate arrays (FPGA) to provide hardware-accelerated implementations of genome analysis algorithms, reducing processing time for a 40× whole human genome from over 8 hours with traditional open-source solutions to approximately 34 minutes.

The engineering behind DRAGEN's performance is also quite interesting. The NovaSeq X Plus features dual socket AMD EPYC 7552 processors with 96 x86_64 cores, four Xilinx Alveo U250 FPGA cards, and 1.5 TB of RAM, enabling on-instrument DRAGEN processing to output fully processed 30× WGS samples approximately once every four minutes while drawing power from a single standard 200-240 volt AC plug (about the same energy required to propel a Tesla Model 3 half a kilometer). The resulting compute cost averages <10 US cents.

However, DRAGEN's impressive technical capabilities come with significant licensing costs. Since Illumina doesn't benefit from FPGA hardware sales, they've adopted an aggressive software licensing model. When running DRAGEN on cloud platforms like AWS, the license cost represents approximately 80% of the total computational expense, a pricing structure that can make cloud-based processing prohibitively expensive for large-scale projects.

Oxford Nanopore and PacBio Revio: Onboard GPUs right inside the device

Oxford Nanopore takes a different approach by building GPUs directly into its high-throughput sequencers, so there’s no need for separate compute clusters or extra software to process the data.

The PromethION P48 now comes with four NVIDIA A100 GPUs, which speed up basecalling by more than 50% compared to older models. This means the sequencer outputs already basecalled data. You get usable FASTQ files right off the machine, ready for analysis. Each run can generate up to 10-13 terabases of data over 72 hours. And because the data is processed in real time on the device, there's no need to move massive raw signal files (which can be 2-3X larger) to the cloud or external servers.

Similarly, PacBio’s Revio also includes onboard NVIDIA A100 GPUs. The system handles basecalling, error correction, and consensus (HiFi) reads on-device, delivering BAM/FASTQ files ready for analysis.

Each SMRT Cell produces ~120 gigabases in ~24 hours. With four cells in parallel, that's up to 480 Gb/day, enough for 1,300 human genomes per year at 30× coverage, all for under $500 per HiFi genome using their SPRQ chemistry and reagent pricing.

Element cloud-first strategy

Element's AVITI platform integrates natively with Amazon Omics, streaming data directly to customer S3 buckets during sequencing and enabling immediate parallel analysis without costly subscription services or intermediate data transfers. Bases2Fastq Ready2Run workflow on Amazon Omics costs approximately $3 to generate FASTQ files for a 300‑G base run (2×150 bp) and completes in about 1 hour for demultiplexing, with the full pipeline (FASTQ → BAM → VCF) finishing in ~2 hours.

Ultima’s hybrid approach

Each UG 100 sequencer comes with built-in high-performance compute that runs standard pipelines like basecalling, alignment, and variant calling directly on the instrument. For downstream analysis, users can upload the output to the cloud and run Ready2Run pipelines on AWS HealthOmics, including DeepVariant ($2 per genome in US regions), Sentieon DNAscope, and GATK workflows. For advanced users, Ultima offers open-source Terraform and WDL modules that make it easy to deploy private workflows using their own AWS storage, compute, and custom references.

Software competitiveness will become more important in the future.

As companies integrate with more multi-omics modalities and become more verticalized, building an ecosystem of robust bioinformatics tools and doing so fast will be important. Prime example is Illumina who’s leading this trend by releasing the Integrated Connected Multiomics platform.

The dynamics in the sequencing market mirror the early smartphone era: As genome sequencing costs plummet, the device matters less than the ecosystem of multiomics kits and tools around it.

Platforms that simplify the transition from raw data to insight will ultimately win the loyalty of users long-term.

Multi-omics Support: Are sequencing machines evolving into something more?

What began as instruments designed to read DNA has quietly transformed into something far more ambitious: platforms that can simultaneously peek into multiple layers of cellular life. Instrument providers seem to be asking a new question: not only "how fast can we sequence DNA?" but also "how completely can we understand a cell?"

This paradigm shift became clear in February 2025 when Illumina announced their spatial technology program. Rather than just reading sequences, their platform maps entire transcriptomes while preserving the spatial context, profiling millions of cells on two slides, supported by Connected Multiomics software designed for researchers without bioinformatics expertise.

Singular had already signaled this direction a year earlier with their G4X announcement in February 2024. Their instrument handles both traditional sequencing and spatial analysis seamlessly. Through Direct-Seq technology, they read RNA directly within intact FFPE tissue without library preparation, simultaneously detecting transcripts, imaging proteins, and reading mutations while maintaining spatial context. Early users have demonstrated 3D tissue reconstructions analyzing over 6 million cells from a single flow cell.

Element also pushed this concept with their AVITI24, performing what they call "5D multiomics": simultaneous detection of DNA, RNA, proteins, phosphorylated proteins, and cell morphology from the same sample in under 24 hours. Their upcoming Direct In Sample Sequencing eliminates library prep entirely by targeting RNA molecules directly within cells. Teton assays focus on specific pathways, currently targeting 350 genes and 50 proteins in MAPK signaling.

PacBio completed this picture with their SPRQ chemistry, launched late 2024. Using 6mA markers in Fiber-seq, every run becomes multiomics by default, simultaneously calling DNA sequence, methylation, and chromatin accessibility. Researchers can extract four molecular views (genome, methylome, epigenome, and transcriptome) from a single sample.

Instrument makers are racing to integrate upstream prep and downstream analysis directly into their platforms through aggressive acquisitions and internal development. Illumina acquired Fluent BioSciences mid 2024 to bring single-cell library prep in-house, then followed with a $425M acquisition of SomaLogic mid 2025 to add proteomics assays to its sequencing ecosystem. Thermo Fisher made a similar move by acquiring Olink for $3.1B late 2023, expanding into high-throughput protein biomarker analysis. Meanwhile, Complete Genomics has internally developed a full-stack system that includes its DNBSEQ sequencing platforms, MGIFLP library prep systems, and a growing multi-omics portfolio spanning spatial transcriptomics, methylation, whole-genome sequencing, single-cell, exome, and proteomics applications.

Starting as DNA readers, these instruments increasingly look more like a “cellular omniscope”, aiming to capture multiple modalities of a cell’s biology.

Read Length

Different applications demand different read lengths. The table below summarizes recommended configurations and throughput per sample for common short-read workflows:

Most short-read sequencing platforms today use a 2x150bp format, which means they sequence 150 bases from each end of a DNA fragment. It also aligns well with common DNA fragment sizes used in WGS libraries, which typically range from 350 to 400 bp. Leading high-throughput instruments, including Illumina's NovaSeq X, Complete Genomics' DNBSEQ-T7/T1+, and GeneMind's SURFSeq Q, all use this 2×150bp configuration.

Mid-throughput instruments like the NextSeq 2000/1000 and MiSeq i100 can support 2×300bp reads, but this extended read length comes at the cost of reduced throughput. This trade-off stems from fundamental limitations in Illumina's sequencing-by-synthesis (SBS) chemistry and bridge amplification process.

Illumina's bridge amplification requires both ends of a DNA fragment to attach to the flow cell surface, forming a bridge structure. The DNA then undergoes in-place amplification to create clusters of identical fragments. This process works optimally with shorter DNA fragments because longer fragments struggle to achieve the necessary flexibility for both ends to efficiently bend and hybridize to the surface. When bridging fails, fragments cannot amplify into clusters, resulting in reduced cluster density, weaker signal intensity, and lower overall yield.

On the contrary, new players in the field such as Roche are breaking the trend with their new SBX system that’s in R&D. Roche promised in their AGBT presentation with “simplex reads ranging from 50 bp to >1,000 bp”, “duplex reads (~150–350 bp inserts) delivering intramolecular consensus with Q39 accuracy”, and even “midi reads reaching 1,200-1,500 bp for longer contiguous coverage. SBX also claims it can sequence 7 human genomes (30× depth) in just one hour, over 5 billion duplex reads per hour. These newer players are likely putting competitive market pressure on Illumina as they are likely bringing the best from both worlds: supporting longer fragments, and while also enabling high-throughput sequencing.

Parting Thoughts

The sequencing landscape in 2025 reflects several converging trends.

Sequencing providers are rapidly verticalizing, transforming from instrument vendors into “full-stack” companies that control everything from sample prep, assays, to data interpretation and insights.

The traditional boundaries between short and long reads continue to blur. Short-read platforms now generate synthetic long reads through molecular barcoding, while long-read companies launch dedicated short-read instruments. Mid-length chemistries promise to bridge the gap with kilobase reads at short-read accuracy levels.

The relentless drive toward lower cost per gigabase shows no signs of slowing. The $100 genome milestone is within reach, with Ultima targeting even lower thresholds potentially approaching $50 per genome by the late 2020s. As reagent costs become commoditized, ecosystem support may emerge as the primary differentiator over raw sequencing economics.

Long-read technologies are poised for clinical adoption as accuracy improvements and cost reductions to <$1000 per genome make long-range genomic profiling possible. The vision of single-platform workflows capturing SNPs, structural variants, methylation, and RNA expression simultaneously is materializing through providers' push for better enhanced chemistry and error correction algorithms.

Finally, sequencers are becoming multiomics platforms. These instruments now capture DNA sequence alongside spatial context, protein abundance, and epigenetic modifications in single runs.

All roads still lead to sequencing machines. But these machines are becoming far more powerful than their original architects ever imagined.

• A 30M cell observational atlas spanning 150 diseases, 200 tissues and 27 technologies

• A partnership with expert data curators Pythia Biosciences and Miraomics

• An agentic human-in-the-loop Python framework for mass scRNA-seq curation

We’re approaching a new era of biology too complex to navigate with unaided human cognition. The industry needs large volumes of structured molecular data to develop a new class of foundation models and build large atlases. It is becoming clear that data is the oil of modern biotech.

Latch is addressing this need by building curation infrastructure, tools and delivery portals for data solution providers. By equipping our partners with powerful new tools, our ambition is to organize the public molecular information scattered across the Internet and provide it for immediate download on a usage basis.

A 30 Million Single Cell Atlas

The need for large scale data has been met initially by purpose-built industrial data-generation projects. These efforts are incredible but do not sample sufficiently broad observational space, especially for rare indications.

Public datasets remain the largest and most diverse reservoir of diseases, tissues and patients. For indications with small patient populations or for complex diseases demanding fine-grained stratification, statistical models must draw on these niche biological states to achieve translational utility.

Our first 30M cell atlas represents thousands of hours of human labor - the combined efforts of the Pythia Biosciences, Miraomics and LatchBio curation teams. It spans broad space of relevant biology, with 150 diseases, 200 tissues, 27 technologies.

This atlas is available in a public portal. Customers can search by study title, abstract or controlled ontology terms and immediately download their data with transparent pricing.

A Partnership with Data Focused Solution Providers

LatchBio is a platform for solution providers. Our team builds compute infrastructure, tools, delivery portals and white labeling features so our customers can focus on their strengths: developing new assays, kits and services.

Our partnership with Miraomics and Pythia Biosciences expands our focus to data delivery. They leverage their expertise in data curation. We provide them with new technology to increase curation throughput, quality and distribute their services to more customers.

An Agentic Human-in-the-Loop Framework for Curation

Recent advances in foundation models and agentic workflows show promise in autonomous scientific reasoning and software development. We hypothesized the structure of the curation problem was particularly suited for these emerging technologies.

We’re also introducing latch-curate, an agentic Python framework that guides an expert scientist through an ordered, step-by-step curation lifecycle and helps them perform tasks like count matrix construction, cell typing and metadata harmonization with greater efficiency and accuracy.

A detailed description of the system design and function can be found in this whitepaper. Reach out to our team for access to this framework.

Our team is building a lot of interesting product at the intersection of molecular data curation and delivery. If you are a solution provider interested in trying our new tools or hosting your data on our portal, please reach out to kenny@latch.bio