Biomolecules · arXiv:2611.17015 v1 · submitted 12 Nov 2026 · 137 references
AI for Biologists: The Knowledge Boundaries of Foundation Models in the Age of Agentic Systems
The paper says
“Artificial intelligence (AI) is changing how biological research is done. Biological foundation models trained on millions of nucleotides, protein, and single-cell measurements predict protein structures, mutation effects, and cellular responses to genetic perturbations.”
more of the abstract
Agentic systems, AI assistants built on large language models, can run multistep analyses from plain-language requests and increasingly call foundation models as tools. Yet it is often unclear when these models' outputs can be trusted. We frame this as a model's knowledge boundary: the range of inputs, tasks, and conditions over which its outputs are reliable. This review explains how foundation models are built and proposes that five design elements (data, architecture, training scheme, evaluation scheme, and post-analysis) define this boundary. We examine evidence on where current models succeed or fail and present standardized descriptions of model limits so that researchers, and the agents acting for them, can use these tools reliably.
from the abstract
Pith says
Machine review · deepseek-v4-flash · 13 Nov 2026 · what was read was not recorded · not checked by a person
It can be wrong. Challenge this review or reply to it.
Holds up, with conditions. A genuinely useful, balanced review of where biological foundation models fail, wrapped around a five-element framework that is asserted rather than demonstrated, and whose proposed boundary statements would not have caught the paper's own failure examples. This line is wrong
2 major objections · 4 minor
How often objections from reviews like this one are right: not yet measured
Not checked: figures, code, what the 137 references say.
The paper says, at Abstract; §2.1.2; §4.3; §6.2
The review did not quote this passage.
A classifier read this sentence of the paper as acknowledging the concern (probability 0.63):
“Few models report whether they can signal that a query lies outside their boundary.”
Pith says: Objection 1 of 2 · major
The stress-test concern about the sufficiency of the five design elements lands. The paper's central claim—that the five elements define the knowledge boundary—is operationalized in Section 6.2 as a machine-readable statement that would let an agent 'confirm that a query lies within a model's training and evaluation domain before relying on its output.'
more of objection 1
But Section 4.3 documents failures on queries that are inside the declared domain: AlphaMissense discordance across 5,845 clinically curated variants in 59 disease genes (ref 118); RNA language models that fail to beat a one-hot encoder for secondary-structure prediction once homologous sequences are excluded (ref 120); and AlphaFold2's failure to detect structural disruption from destabilizing point mutations (ref 100). The paper's own motivating case, Paper2Agent using AlphaGenome for regulatory variant scoring at 1p13 (b230–b233), produced a wrong-gene ranking and a wrong-direction expression claim while operating on the model's intended task. A statement listing training corpus, tokenization, context length, training objective, benchmarks, and post-analysis methods would not have marked any of these queries as out-of-boundary. The paper partially acknowledges the gap ('Benchmark performance, confidence scores, transferable embeddings, and reliability in application must therefore be established separately,' end of Section 4.3; b326), but Section 6.2 does not incorporate these caveats into the proposed mechanism. The authors should either weaken the claim from 'define' to 'organize and partially determine,' explicitly distinguishing the declared domain from the verified reliable range, or specify the additional components (evaluation-coverage records, out-of-distribution or calibration signals, explicit failure-mode lists) that would make the statement support prospective reliability judgments, and demonstrate on at least one documented case what the statement would have flagged.
The paper says, at §3 (b225–b227) vs §5.1 (ref 123)
The review did not quote this passage.
Pith says: Objection 2 of 2 · major
The paper presents BioReason's results as evidence for the general principle that 'when [the answer] lies beyond [the LLM's textual knowledge], the encoder sets the ceiling of knowledge' (b227), citing the improvement from 49% to 80% accuracy after adding Evo 2 (b226).
more of objection 2
Section 5.1, however, cites ref 123, which found that BioReason and BioReason-Pro were as accurate with shuffled biological inputs as with intact ones. The paper should reconcile these two statements: either qualify the Section 3 claim as task-dependent, or explain why the shuffled-input control does not undermine the attribution of the gain to the encoder. As written, the paper's most concrete example of an agent inheriting a foundation model's boundary is in tension with a control study it cites.
4 minor objections
The paper says, at §3
Pith says: minor objection 1 of 4
The text 'runs locally, or at remove facilities without internet access' contains a typo: 'remove' should read 'remote.'
The paper says, at §6.2; Data and Code Availability
Pith says: minor objection 2 of 4
The online resource URL appears as the placeholder '[URL]' in both Section 6.2 and the Data and Code Availability section; this should be resolved before publication.
show 2 more
The paper says, at References (77, 84, 89)
Pith says: minor objection 3 of 4
Several reference entries lack identifiers (e.g., ref 77, the Anthropic Engineering Blog post; ref 84, the Biomni-R0 technical report; ref 89, the OpenAI mathematics statement), which will complicate verification; the reference list should be brought to journal style.
The paper says, at Abstract; §6.2
Pith says: minor objection 4 of 4
The relationship between 'training and evaluation domain' and 'knowledge boundary' should be defined explicitly, since the abstract defines the boundary as the range over which outputs are reliable whereas Section 6.2 uses 'domain' to mean what the model was exposed to; these are not the same set, as the Section 4.3 evidence shows.
What would settle it
Construct, from Section 4.3, the five-element boundary statement that Section 6.2's template would generate for each failing model, AlphaMissense on ref 118, RNA models on ref 120, AlphaFold2 on ref 100, and AlphaGenome on the SORT1 case (b230-b233). For each, determine whether the query lies inside the declared boundary, training corpus, tokenization, context length, objective, benchmark list, post-analysis. If any query falls fully inside the declared boundary yet the model output is unreliable, then a boundary statement alone cannot deliver the promised reliability decision and Section 6.2 requires an additional reliability-calibrating component (e.g., per-task error rates or calibrated confidence). If every failure query falls outside its statement's declared boundary, the framework as specified is sufficient and this concern is resolved.
from the skeptic's reading
How the review went
4 readings by deepseek-v4-flash · 13 Nov 2026
- Reader holds up, with conditions
- Skeptic partly agrees
- Editor wrote the verdict above
- Referee major revision: 2 major objections, 4 minor
Not run on this paper: the simulated authors' defence.
Pith did not check this review's objections against the paper.
The review in full
The review in brief, machine-written
Worth reading, and worth sending to referees. The paper's real asset is its evidence: it collects the independent baseline studies that cut against biological foundation models, including ProteinGym's moderate unsupervised performance, linear models matching perturbation predictors, parameter-free single-cell representations, AlphaMissense discordance on clinical variants, RNA language models no better than one-hot once homologs are excluded, and AlphaFold2 missing destabilizing mutations. It also includes the SORT1/Paper2Agent case, where an agent used AlphaGenome for its intended task and reached the right gene only by contradicting both the tool's output and the experimental literature. Few reviews in this space are this honest. What is actually new is smaller. The five design elements are a reasonable organizing template, and knowledge boundary is a useful label, but the components are known and the authors explicitly build on model cards and datasheets. There is no new data, no systematic search, and no delivered specification. The central soft spot is the leap from design elements to prospective trust. Section 6.2 proposes machine-readable boundary statements that would let an agent confirm a query lies within a model's training and evaluation domain. But Section 4.3's failures are inside that domain. AlphaMissense failed on curated clinical variants in disease genes; the RNA models failed once homologous sequences were removed; AlphaFold2 largely missed destabilizing point mutations. A statement recording corpus, architecture, objective, benchmarks, and post-analysis would not flag any of those queries as out-of-boundary. The SORT1 example is the cleanest version: the model was used on its own task and gave a wrong-gene, wrong-direction answer. I checked the stress-test concern and it lands. A statement of where a model was tested is not a certificate of reliability for a new query, and listing evaluation benchmarks is retrospective. The paper frames this as a perspective and acknowledges that few models can signal out-of-boundary queries, but the proposed deliverable needs either a concrete schema plus empirical test, or a more modest claim. Minor issues: no actual boundary-statement template or worked example is given, the companion resource URL is a placeholder, and the literature selection is not systematic, though it looks balanced. This deserves a serious referee. Send it out, and ask referees to focus on whether boundary statements can support prospective decisions and to require at least one worked example. For anyone working in this area, it is worth citing for its landscape and its baseline evidence.
The strongest objection
The central deliverable (Section 6.2) is a machine-readable boundary statement, organized by the five design elements, that would let an agent "confirm that a query lies within a model's training and evaluation domain before relying on its output." For this to work, the statement must support a prospective reliability judgment. The paper asserts this (b369) but does not test it, and its own evidence cuts against the sufficiency claim in two ways. First, within-domain failures: Section 4.3 documents cases where models fail precisely on queries their design elements cover, AlphaMissense discordance across 5,845 clinically curated variants in 59 disease genes (ref 118), RNA language models no better than a one-hot encoder once homologous sequences are excluded (ref 120), and AlphaFold2 largely missing destabilizing point mutations (ref 100). A statement listing training corpus, tokenization, context length, objective, benchmarks, and post-analysis would not flag any of these as out-of-boundary: they sit inside the data distribution, within architecture limits, and on the models' own tasks. Second, the Paper2Agent/SORT1 example (b230-b233), the paper's own motivating case, is a within-boundary confident failure: AlphaGenome was used for its intended task (regulatory variant scoring) and produced a wrong-gene, wrong-direction output; no boundary statement would have intercepted it. Two consequences follow. (a) A statement that records the evaluation domain cannot distinguish reliable from unreliable queries inside that domain, yet that distinction is exactly the decision the agent must make. (b) The framework risks circularity: evaluation scheme is both one of the five defining elements (Section 4.1) and the instrument that reveals the boundary, so listing benchmarks "used" (Section 6.2) is retrospective, not predictive. The leap from "design elements shape capabilities" to "a statement of them supports prospective trust" is the unverified premise.
The reader's CONDITIONAL verdict is the right one, and my analysis does not move it. My concern is close to, but narrower than, the reader's weakest_assumption. The reader emphasized that reliability may be task- and context-dependent in ways a fixed boundary statement cannot encode. I sharpen this: even with the task held fixed, the paper's own Section 4.3 supplies documented within-domain reliability failures (AlphaMissense on curated clinical variants; RNA models on homology-reduced tests; AlphaFold2 on destabilizing point mutations), and the SORT1 case shows a confident within-domain wrong answer that a boundary statement recording training/evaluation domain would not flag. So the unverified premise is not only task-dependence but the very sufficiency of a static statement organized by the five elements to support a prospective reliability decision. This is a genuine soft spot in the central claim that boundary statements let agents and researchers rely on models, but the paper is a perspective proposing new infrastructure, the authors explicitly acknowledge the difficulty of knowing a model's boundaries (b83, b253, b326), and the gap is addressable by adding reliability-calibrating content to the statement template. That supports keeping the verdict CONDITIONAL rather than ACCEPT, and does not justify REJECT. I do not regard the inaccessible companion resource or the absence of delivered standardized descriptions as load-bearing for the scientific claim; those are secondary. I agree with the reader's overall posture and verdict, differing only in the precise location of the weak premise.
The referee's report
The review is timely and useful as an organizing map of a fast-moving field. Its principal strength is that it does not rest on developer-reported results: it foregrounds independent reassessments and simple-baseline comparisons (e.g., refs 100, 102–117, 118–120) that cut against model claims, and it is candid about the field's early stage and about circularity risks in LLM-based annotation (b288). The five-element taxonomy is a clear pedagogical device, and the proposal to attach knowledge-boundary statements to model releases (Section 6.2) addresses a real operational need for both human users and agents. If the sufficiency claim were validated, the paper would give the field a standard vocabulary for model limits. However, the central claim is asserted rather than demonstrated, and the paper's own Section 4.3 evidence includes failures inside the declared domain that a static five-element statement would not flag; the significance is therefore conditional on the authors reframing or defending the sufficiency claim.
Pith's reading of the paper (a machine's summary, not the authors')
This review argues that a biological foundation model's reliability is not open-ended but is set by five choices made when the model is built: data, architecture, training scheme, evaluation scheme, and post-analysis. The authors call the resulting range of inputs, tasks, and conditions the model's knowledge boundary, and they propose that every model be released with a structured, machine-readable boundary statement organized by those five elements. The motivation is that agentic systems increasingly call foundation models as tools and inherit whatever boundary the foundation model has, so errors from outside that boundary can look as authoritative as well-supported results. The review supports the framework with published evidence on where models succeed and where simple baselines match or beat them, concluding that making boundaries explicit is what would let biologists and agents use these tools reliably.
What carries the argument. The organizing object is the knowledge boundary, the range of inputs, tasks, and conditions over which a model's outputs are reliable, and the carrying mechanism is a five-element template used to decompose that boundary: data, architecture, training scheme, evaluation scheme, and post-analysis. Each element is tied to a specific kind of failure: data limits what can be known, architecture limits what relationships can be captured, the training objective determines what is rewarded, evaluation determines which weaknesses become visible, and post-analysis determines what knowledge becomes accessible. The template does the argument's work by turning an abstract trust question into a concrete checklist, by tracing observed failures back to design choices, and by providing the schema for the proposed machine-readable boundary statements that agents could consult before calling a model.
Core claim. The paper's central claim is that a biological foundation model's knowledge boundary, defined as the range of inputs, tasks, and conditions over which its outputs can be relied upon, is set by five design elements: the data and tokenization, the model architecture, the training scheme, the evaluation scheme, and the post-analysis methods such as probing, fine-tuning, zero-shot scoring, and interpretability. The review assembles two bodies of evidence for this claim: studies showing that self-supervised models capture real biological structure, from residue chemistry and contacts to evolutionary constraint and functional design, and independent studies showing that the same models stop being reliable where simple linear models, mean responses, or one-hot encodings match or beat them, especially for perturbation prediction, single-cell tasks, and mutation-dense design. From this, the authors conclude that integrating foundation models with agentic systems depends on making the knowledge boundary explicit through structured boundary statements that record each model's five elements in a machine-readable form, in the spirit of model cards and datasheets.
The premise it rests on. The framework assumes that a model's reliability can be captured by a fixed description of its five design elements that holds across tasks, users, and queries, and the paper asserts this rather than testing it.
If the paper is right (Pith's inferences, not the authors')
If boundary statements become standard, an agent can route each question to the model whose stated boundary covers it and fall back to a simple baseline when no model applies.
Benchmark results would be reported together with simple baselines and train-test similarity controls, making it harder for a high score to reflect recall of near-duplicate training sequences rather than learned rules.
The field would have a principled reason to develop specialized models for boundary regions where general models fail, rather than waiting for one ever-larger general-purpose model.
New data collection, including self-driving laboratory experiments, could be targeted directly at the regions a boundary statement shows are missing.
A correct final answer from an agent would no longer be treated as evidence that the underlying model reasoning was sound, because the boundary statement would say which post-analysis conditions the result depends on.
Editorial inference: the same five-element template could be applied to the large language models inside agentic systems themselves, since the planner has its own training data, architecture, objective, evaluation, and post-processing boundaries.
Editorial inference: if boundary statements became the norm, the unit of scientific reproducibility could shift from the model weights alone to the model-plus-boundary-statement package, making negative results and attribution of failures easier to record.
Editorial inference: a directly testable extension is a routing benchmark in which agents receive machine-readable boundary statements for a portfolio of models and are scored on choosing the right model for out-of-distribution queries; the paper argues for such routing but does not supply the experiment.
Editorial inference: the reviewed evidence implies that publishing only average benchmark scores hides most failures, and that reporting failure distributions conditioned on sequence similarity and task type would be a cheap near-term implementation of the proposal.
Check this review
Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.
How often reviews like this one are right
Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.
How often objections from reviews like this one are right: not yet measured
Reviews by people
This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.
The paper
Abstract
Artificial intelligence (AI) is changing how biological research is done. Biological foundation models trained on millions of nucleotides, protein, and single-cell measurements predict protein structures, mutation effects, and cellular responses to genetic perturbations. Agentic systems, AI assistants built on large language models, can run multistep analyses from plain-language requests and increasingly call foundation models as tools. Yet it is often unclear when these models' outputs can be trusted. We frame this as a model's knowledge boundary: the range of inputs, tasks, and conditions over which its outputs are reliable. This review explains how foundation models are built and proposes that five design elements (data, architecture, training scheme, evaluation scheme, and post-analysis) define this boundary. We examine evidence on where current models succeed or fail and present standardized descriptions of model limits so that researchers, and the agents acting for them, can use these tools reliably.
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.
abstract compiled · 2026-11-13
The authors' main claim, as the classifier picked it out 0.63
“We examine evidence on where current models succeed or fail and present standardized descriptions of model limits so that researchers, and the agents acting for them, can use these tools reliably.”
How the main claim is stated: asserted 0.97
What the abstract itself states, as probabilities: claims first 0.04 · states limitation 0.13 · human participants 0.07 · controlled comparison 0.03 · randomized assignment 0.01 · states quantitative result 0.05 · compares with prior results 0.07 · mentions code or data availability 0.03
The abstract, sentence by sentence
This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.
| Sentence | Role | The authors' own work | States the main conclusion | Strength |
|---|---|---|---|---|
| Artificial intelligence (AI) is changing how biological research is done. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| Biological foundation models trained on millions of nucleotides, protein, and single-cell measurements predict protein structures, mutation effects, and cellular responses to genetic perturbations. | background 1.00 | 0.07 | 0.05 | asserted 1.00 |
| Agentic systems, AI assistants built on large language models, can run multistep analyses from plain-language requests and increasingly call foundation models as tools. | background 1.00 | 0.06 | 0.03 | asserted 0.92 |
| Yet it is often unclear when these models' outputs can be trusted. | gap 0.85 | 0.08 | 0.04 | asserted 0.79 |
| We frame this as a model's knowledge boundary: the range of inputs, tasks, and conditions over which its outputs are reliable. | contribution 0.84 | 0.67 | 0.14 | asserted 0.99 |
| This review explains how foundation models are built and proposes that five design elements (data, architecture, training scheme, evaluation scheme, and post-analysis) define this boundary. | contribution 0.99 | 0.73 | 0.30 | asserted 0.46 |
| We examine evidence on where current models succeed or fail and present standardized descriptions of model limits so that researchers, and the agents acting for them, can use these tools reliably. | contribution 0.98 | 0.73 | 0.49 | asserted 0.97 |
What kind of paper the classifier reports this is: review 1.00 · evidence from literature 0.99
What was examined, and what was not
Sources read
- arXiv listing abstract v1
Not examined
- full text
- figures and tables
- references and citing sentences
- relations to other papers
Levels run
- P abstract
Not run
- S sections
- T body sentences
- C citations
- R cross-paper relations
How the labels were made
Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-13: 1 request, 5,520 input tokens.
Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON
Figures (3)
References (137)
- [1] Kendrew JC, Dickerson RE, Strandberg BE, Hart RG, Davies DR, et al. 1960. Structure of myoglobin: a three-dimensional Fourier synthesis at 2 Å resolution. Nature 185: 422- 27 on Pith
- [2] Berman HM, Westbrook J, Feng Z, Gilliland G, Bhat TN, et al. 2000. The Protein Data Bank. Nucleic Acids Res. 28: 235-42 39 on Pith
- [3] wwPDB (Worldwide Protein Data Bank). 2026. PDB reaches a new milestone. EMBL- EBI News on Pith
- [4] Jumper J, Evans R, Pritzel A, Green T, Figurnov M, et al. 2021. Highly accurate protein structure prediction with AlphaFold. Nature 596: 583-89 on Pith
- [5] Abramson J, Adler J, Dunger J, Evans R, Green T, et al. 2024. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630: 493-500 on Pith
- [6] Lin Z, Akin H, Rao R, Hie B, Zhu Z, et al. 2023. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379: 1123-30 on Pith
- [7] Dalla-Torre H, Gonzalez L, Mendoza-Revilla J, Lopez Carranza N, Grzywaczewski AH, et al. 2025. Nucleotide Transformer: building and evaluating robust foundation models for human genomics. Nat. Methods 22: 287-97 on Pith
- [8] Brixi G, Durrant MG, Ku J, Naghipourfar M, Poli M, et al. 2026. Genome modelling and design across all domains of life with Evo 2. Nature 652: 1349-61 on Pith
- [9] Theodoris CV, Xiao L, Chopra A, Chaffin MD, Al Sayed ZR, et al. 2023. Transfer learning enables predictions in network biology. Nature 618: 616-24 on Pith
- [10] Cui H, Wang C, Maan H, Pang K, Luo F, et al. 2024. scGPT: toward building a foundation model for single-cell multi-omics using generative AI. Nat. Methods 21: 1470-80 on Pith
- [11] Rosen Y, Roohani Y, Agrawal A, Samotorcan L, Tabula Sapiens Consortium, et al. 2023. Universal Cell Embeddings: a foundation model for cell biology. bioRxiv 2023.11.28.568918 on Pith
- [12] Wang Z, Jin Q, Wei CH, Tian S, Lai PT, et al. 2025. GeneAgent: self-verification language agent for gene-set analysis using domain databases. Nat. Methods 22: 1677-85 on Pith
- [13] Alber S, Chen B, Sun E, Isakova A, Wilk AJ, Zou J. 2026. CellVoyager: AI CompBio agent generates new insights by autonomously analyzing biological data. Nat. Methods 23: 749-59 on Pith
- [14] Gao S, Zhu R, Kong Z, Noori A, Su X, et al. 2025. TxAgent: an AI agent for therapeutic reasoning across a universe of tools. arXiv 2503.10970 [cs.AI] on PithPith's reviewarXiv
- [15] Mitchener L, Laurent JM, Tenmann B, Narayanan S, Wellawatte GP, et al. 2025. BixBench: a comprehensive benchmark for LLM-based agents in computational biology. arXiv 2503.00096 [q-bio.QM] on PitharXiv
- [16] Panigrahi SS, Videnović J, Brbić M. 2026. HeurekaBench: a benchmarking framework for AI co-scientist. arXiv 2601.01678 [cs.AI] on PitharXiv
- [17] Huang K, Zhang S, Wang H, Qu Y, Lu Y, et al. 2026. Autonomous biomedical research with an artificial intelligence agent. Science on Pith
- [18] Zhang HG, Eckmann P, Miao J, Mahon AB, Zou J. 2026. The Virtual Biotech: a multi- agent AI framework for therapeutic discovery and development. bioRxiv 2026.02.23.707551 on Pith
- [19] de Almeida BP, Richard G, Dalla-Torre H, Blum C, Hexemer L, et al. 2025. A multimodal conversational agent for DNA, RNA and protein tasks. Nat. Mach. Intell. 7: 928–41 on Pith
- [20] Schaefer M, Peneder P, Malzl D, Lombardo SD, Peycheva M, et al. 2026. Multimodal learning enables chat-based exploration of single-cell data. Nat. Biotechnol. 44: 1722–32 on Pith
show 80 more
- [21] Fallahpour A, Magnuson A, Gupta P, Ma S, Naimer J, et al. 2025. BioReason: incentivizing multimodal biological reasoning within a DNA-LLM model. In Advances in Neural Information Processing Systems 38 (NeurIPS 2025) 40 on Pith
- [22] Bommasani R, Hudson DA, Adeli E, Altman R, Arora S, et al. 2021. On the opportunities and risks of foundation models. arXiv 2108.07258 [cs.LG] on PithPith's reviewarXiv
- [23] Avsec Ž, Latysheva N, Cheng J, Novati G, Taylor KR, et al. 2026. Advancing regulatory variant effect prediction with AlphaGenome. Nature 649: 1206-18 on Pith
- [24] Adduri AK, Gautam D, Bevilacqua B, Imran A, Shah R, et al. 2025. Predicting cellular responses to perturbation across diverse contexts with State. bioRxiv 2025.06.26.661135 on Pith
- [25] Meier J, Rao R, Verkuil R, Liu J, Sercu T, Rives A. 2021. Language models enable zero- shot prediction of the effects of mutations on protein function. In Advances in Neural Information Processing Systems 34 (NeurIPS 2021), pp. 29287-303. Red Hook, NY: Curran on Pith
- [26] Zhou Z, Ji Y, Li W, Dutta P, Davuluri R, Liu H. 2024. DNABERT-2: efficient foundation model and benchmark for multi-species genome. In 12th International Conference on Learning Representations (ICLR 2024). Vienna on Pith
- [27] Sanabria M, Hirsch J, Joubert PM, Poetsch AR. 2024. DNA language model GROVER learns sequence context in the human genome. Nat. Mach. Intell. 6: 911-23 on Pith
- [28] UniProt Consortium. 2025. UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Res. 53: D609–17 on Pith
- [29] CZI Cell Science Program, Abdulla S, Aevermann B, Assis P, Badajoz S, et al. 2025. CZ CELLxGENE Discover: a single-cell data platform for scalable exploration, analysis and modeling of aggregated data. Nucleic Acids Res. 53: D886–900 on Pith
- [30] Hayes T, Rao R, Akin H, Sofroniew NJ, Oktay D, et al. 2025. Simulating 500 million years of evolution with a language model. Science 387: 850-58 on Pith
- [31] Grattafiori A, Dubey A, Jauhri A, Pandey A, Kadian A, et al. 2024. The Llama 3 herd of models. arXiv 2407.21783 [cs.AI] on PithPith's reviewarXiv
- [32] Ding F, Steinhardt J. 2024. Protein language models are biased by unequal sequence sampling across the tree of life. bioRxiv 2024.03.07.584001 on Pith
- [33] Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I. 2005. Invariant visual representation by single neurons in the human brain. Nature 435: 1102-07 on Pith
- [34] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, et al. 2017. Attention is all you need. In Advances in Neural Information Processing Systems 30 (NIPS 2017), pp. 5998- on Pith
- [35] Nguyen E, Poli M, Faizi M, Thomas A, Wornow M, et al. 2023. HyenaDNA: long-range genomic sequence modeling at single nucleotide resolution. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Red Hook, NY: Curran on Pith
- [36] Schiff Y, Kao CH, Gokaslan A, Dao T, Gu A, Kuleshov V. 2024. Caduceus: bi- directional equivariant long-range DNA sequence modeling. Proc. Mach. Learn. Res. 235: 43632-48 on Pith
- [37] Li MM, Huang Y, Sumathipala M, Liang MQ, Valdeolivas A, et al. 2024. Contextual AI models for single-cell protein biology. Nat. Methods 21: 1546-57 on Pith
- [38] Devlin J, Chang MW, Lee K, Toutanova K. 2019. BERT: pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp... on Pith
- [39] Radford A, Narasimhan K, Salimans T, Sutskever I. 2018. Improving language understanding by generative pre-training, OpenAI 41 on Pith
- [40] Nijkamp E, Ruffolo JA, Weinstein EN, Naik N, Madani A. 2023. ProGen2: exploring the boundaries of protein language models. Cell Syst. 14: 968-78.e3 on Pith
- [41] Nguyen E, Poli M, Durrant MG, Kang B, Katrekar D, et al. 2024. Sequence modeling and design from molecular to genome scale with Evo. Science 386: eado9336 on Pith
- [42] Fradkin P, Shi R, Isaev K, Frey BJ, Morris Q, et al. 2024. Orthrus: towards evolutionary and functional RNA foundation models. bioRxiv 2024.10.10.617658 on Pith
- [43] Heimberg G, Kuo T, DePianto DJ, Salem O, Heigl T, et al. 2025. A cell atlas foundation model for scalable search of similar human cells. Nature 638: 1085-94 on Pith
- [44] Notin P, Kollasch AW, Ritter D, van Niekerk L, Paul S, et al. 2023. ProteinGym: large- scale benchmarks for protein fitness prediction and design. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). , Datasets and Benchmarks Track. Red Hook, NY: Curran on Pith
- [45] Pampari A, Shcherbina A, Kvon EZ, Kosicki M, Nair S, et al. 2024. ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis- regulatory sequence syntax, transcription factor footprints and regulatory variants. bioRxiv 2024.12.25.630221 on Pith
- [46] Chen J, Hu Z, Sun S, Tan Q, Wang Y, et al. 2022. Interpretable RNA foundation model from unannotated data for highly accurate RNA structure and function predictions. arXiv 2204.00300 [q-bio.QM] on PithPith's reviewarXiv
- [47] Penić RJ, Vlašić T, Huber RG, Wan Y, Šikić M. 2025. RiNALMo: general-purpose RNA language models can generalize well on structure prediction tasks. Nat. Commun. 16: 5671 on Pith
- [48] Kang B, Fan R, Yi M, Cui C, Cui Q. 2025. A large-scale foundation model for bulk transcriptomes. bioRxiv 2025.06.11.659222 on Pith
- [49] Tejada-Lapuerta A, Schaar AC, Gutgesell R, Palla G, Halle L, et al. 2025. Nicheformer: a foundation model for single-cell and spatial omics. Nat. Methods 22: 2525–38 on Pith
- [50] Wenckstern J, Jain E, Vasilev K, Pariset M, Wicki A, et al. 2025. AI-powered virtual tissues from spatial proteomics for clinical diagnostics and biomedical discovery. arXiv 2501.06039 [q-bio.QM] on PitharXiv
- [51] Elnaggar A, Heinzinger M, Dallago C, Rehawi G, Wang Y, et al. 2022. ProtTrans: toward understanding the language of life through self-supervised learning. IEEE Trans. Pattern Anal. Mach. Intell. 44: 7112-27 on Pith
- [52] Su J, Han C, Zhou Y, Shan J, Zhou X, Yuan F. 2024. SaProt: protein language modeling with structure-aware vocabulary. In 12th International Conference on Learning Representations (ICLR 2024). Vienna on Pith
- [53] Elnaggar A, Essam H, Salah-Eldin W, Moustafa W, Elkerdawy M, et al. 2023. Ankh: optimized protein language model unlocks general-purpose modelling. arXiv 2301.06568 [cs.LG] on PithPith's reviewarXiv
- [54] Yang X, Liu G, Feng G, Bu D, Wang P, et al. 2024. GeneCompass: deciphering universal gene regulatory mechanisms with a knowledge-informed cross-species foundation model. Cell Res. 34: 830-45 on Pith
- [55] Song L, Segal E, Xing E. 2024. Toward AI-driven digital organism: multiscale foundation models for predicting, simulating and programming biology at all levels. arXiv 2412.06993 [q-bio.QM] 42 on PithPith's reviewarXiv
- [56] Avsec Ž, Agarwal V, Visentin D, Ledsam JR, Grabska-Barwinska A, et al. 2021. Effective gene expression prediction from sequence by integrating long-range interactions. Nat. Methods 18: 1196-203 on Pith
- [57] Linder J, Srivastava D, Yuan H, Agarwal V, Kelley DR. 2025. Predicting RNA-seq coverage from DNA sequence as a unifying model of gene regulation. Nat. Genet. 57: 949-61 on Pith
- [58] Hingerl JC, Martens LD, Karollus A, Manz T, Buenrostro JD, et al. 2025. scooby: modeling multimodal genomic profiles from DNA sequence at single-cell resolution. Nat. Methods 22: 2275–85 on Pith
- [59] Frazer J, Notin P, Dias M, Gomez A, Min JK, et al. 2021. Disease variant prediction with deep generative models of evolutionary data. Nature 599: 91-95 on Pith
- [60] Cheng J, Novati G, Pan J, Bycroft C, Žemgulytė A, et al. 2023. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381: eadg7492 on Pith
- [61] Orenbuch R, Shearer CA, Kollasch AW, Spinner AD, Hopf T, et al. 2025. Proteome- wide model for human disease genetics. Nat. Genet. 57: 3165–74 on Pith
- [62] Madani A, Krause B, Greene ER, Subramanian S, Mohr BP, et al. 2023. Large language models generate functional protein sequences across diverse families. Nat. Biotechnol. 41: 1099-106 on Pith
- [63] Lal A, Garfield D, Biancalani T, Eraslan G. 2024. Designing realistic regulatory DNA with autoregressive language models. Genome Res. 34: 1411-20 on Pith
- [64] Fallahpour A, Gureghian V, Filion GJ, Lindner AB, Pandi A. 2025. CodonTransformer: a multispecies codon optimizer using context-aware neural networks. Nat. Commun. 16: 3205 on Pith
- [65] Joshi CK, Jamasb AR, Viñas R, Harris C, Mathis SV, et al. 2025. gRNAde: geometric deep learning for 3D RNA inverse design. In 13th International Conference on Learning Representations (ICLR 2025). Singapore on Pith
- [66] Bunne C, Stark SG, Gut G, del Castillo JS, Levesque M, et al. 2023. Learning single-cell perturbation responses using neural optimal transport. Nat. Methods 20: 1759-68 on Pith
- [67] Roohani Y, Huang K, Leskovec J. 2024. Predicting transcriptional outcomes of novel multigene perturbations with GEARS. Nat. Biotechnol. 42: 927-35 on Pith
- [68] Ouyang L, Wu J, Jiang X, Almeida D, Wainwright CL, et al. 2022. Training language models to follow instructions with human feedback. In Advances in Neural Information Processing Systems 35 (NeurIPS 2022), pp. 27730-44. Red Hook, NY: Curran on Pith
- [69] Lewis P, Perez E, Piktus A, Petroni F, Karpukhin V, et al. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems 33 (NeurIPS 2020), pp. 9459-74. Red Hook, NY: Curran on Pith
- [70] Patel J, Downing J, Ferguson H, You T, Malinverni D, et al. 2026. Democratizing reliable knowledge-seeking with MyGPT: a privacy-first, open-source retrieval- augmented generation platform. Nat. Commun. 17 on Pith
- [71] Schick T, Dwivedi-Yu J, Dessì R, Raileanu R, Lomeli M, et al. 2023. Toolformer: language models can teach themselves to use tools. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Red Hook, NY: Curran on Pith
- [72] Yao S, Zhao J, Yu D, Du N, Shafran I, et al. 2023. ReAct: synergizing reasoning and acting in language models. In 11th International Conference on Learning Representations (ICLR 2023), Kigali. Rwanda 43 on Pith
- [73] Guo D, Yang D, Zhang H, Song J, Wang P, et al. 2025. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning. Nature 645: 633-38 on Pith
- [74] Yue Y, Chen Z, Lu R, Zhao A, Wang Z, et al. 2025. Does reinforcement learning really incentivize reasoning capacity in LLMs beyond the base model? arXiv 2504.13837 [cs.AI] on PithPith's reviewarXiv
- [75] Anthropic. 2024. Introducing the Model Context Protocol. Anthropic News on Pith
- [76] Anthropic. 2025. Equipping agents for the real world with Agent Skills. Anthropic Engineering Blog on Pith
- [77] Young J. 2025. Effective harnesses for long-running agents. Anthropic Engineering Blog on Pith
- [78] Miao J, Davis JR, Zhang Y, Pritchard JK, Zou J. 2026. Reimagining research papers as interactive and reliable AI agents. Nature 658: 1272-80 on Pith
- [79] Bran AM, Cox S, Schilter O, Baldassari C, White AD, Schwaller P. 2024. Augmenting large language models with chemistry tools. Nat. Mach. Intell. 6: 525-35 on Pith
- [80] Qu Y, Huang K, Yin M, Zhan K, Liu D, et al. 2026. CRISPR-GPT for agentic automation of gene-editing experiments. Nat. Biomed. Eng. 10: 245-58 on Pith
- [81] Swanson K, Wu W, Bulaong NL, Pak JE, Zou J. 2025. The Virtual Lab of AI agents designs new SARS-CoV-2 nanobodies. Nature 646: 716-23 on Pith
- [82] Gottweis J, Weng WH, Daryin A, Tu T, Sirkovic P, et al. 2026. Accelerating scientific discovery with Co-Scientist. Nature 655: 487-96 on Pith
- [83] Ghareeb AE, Chang B, Mitchener L, Yiu A, Szostkiewicz CJ, et al. 2026. A multi-agent system for automating scientific discovery. Nature 655: 497-505 on Pith
- [84] Biomni Team. 2025. Biomni-R0: agentic LLMs trained end-to-end with multi-turn reinforcement learning for expert-level biomedical reasoning on Pith
- [85] Rizvi SA, Levine D, Patel A, Zhang S, Wang E, et al. 2025. Scaling large language models for next-generation single-cell analysis. bioRxiv 2025.04.14.648850 on Pith
- [86] Istrate AM, Milletari F, Castrotorres F, Tomczak JM, Torkar M, et al. 2025. rbio1 – training scientific reasoning LLMs with biological world models as soft verifiers. bioRxiv 2025.08.18.670981 on Pith
- [87] Turpin M, Michael J, Perez E, Bowman SR. 2023. Language models don't always say what they think: unfaithful explanations in chain-of-thought prompting. In Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Red Hook, NY: Curran on Pith
- [88] Chen Y, Benton J, Radhakrishnan A, Uesato J, Denison C, et al. 2025. Reasoning models don't always say what they think. arXiv 2505.05410 [cs.CL] on PithPith's reviewarXiv
- [89] OpenAI. 2026. Ten advances in mathematics and theoretical computer science on Pith
- [90] Musunuru K, Strong A, Frank-Kamenetsky M, Lee NE, Ahfeldt T, et al. 2010. From noncoding variant to phenotype via SORT1 at the 1p13 cholesterol locus. Nature 466: 714-19 on Pith
- [91] Rives A, Meier J, Sercu T, Goyal S, Lin Z, et al. 2021. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. PNAS 118: e2016239118 on Pith
- [92] Rao R, Meier J, Sercu T, Ovchinnikov S, Rives A. 2021. Transformer protein language models are unsupervised structure learners. In 9th International Conference on Learning Representations (ICLR 2021). virtual on Pith
- [93] Kryshtafovych A, Schwede T, Topf M, Fidelis K, Moult J. 2021. Critical assessment of methods of protein structure prediction (CASP)—Round XIV. Proteins 89: 1607-17 44 on Pith
- [94] Benegas G, Albors C, Aw AJ, Ye C, Song YS. 2025. A DNA language model based on multispecies alignment predicts the effects of genome-wide variants. Nat. Biotechnol. 43: 1960–65 on Pith
- [95] Simon E, Zou J. 2025. InterPLM: discovering interpretable features in protein language models via sparse autoencoders. Nat. Methods 22: 2107-17 on Pith
- [96] Adams E, Bai L, Lee M, Yu Y, AlQuraishi M. 2025. From mechanistic interpretability to mechanistic biology: training, evaluating, and interpreting sparse autoencoders on protein language models. Proc. Mach. Learn. Res. 267 on Pith
- [97] Gujral O, Bafna M, Alm E, Berger B. 2025. Sparse autoencoders uncover biologically interpretable features in protein language model representations. PNAS 122 on Pith
- [98] Hie BL, Shanker VR, Xu D, Bruun TUJ, Weidenbacher PA, et al. 2024. Efficient evolution of human antibodies from general protein language models. Nat. Biotechnol. 42: 275-83 on Pith
- [99] Huang X, Long Q, Xiao M, Chen J, Zhu H, et al. 2026. SciHorizon-Gene: benchmarking LLM for life sciences inference from gene knowledge to functional understanding. arXiv 2601.12805 [cs.CL] on PithPith's reviewarXiv
- [100] Buel GR, Walters KJ. 2022. Can AlphaFold2 predict the impact of missense mutations on structure? Nat. Struct. Mol. Biol. 29: 1-2 on Pith
The first 100 of 137 references are listed here. All of them are in this paper's citation record.
Challenges and discussion
Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.
Come back by email
A short list of the week's newly reviewed papers. Nothing is sent until you confirm from your inbox, and every email has a link that stops it. No account.



No comments yet.
Sign in to comment