REVIEW 5 major objections 6 minor 67 references
SciForge argues that the bottleneck in AI-driven science is not the model but the missing persistent, auditable research state, and that a local-first workbench with evidence graphs and translate-then-reason input can supply it.
Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →
An open-source research workbench that attaches provenance, audit, and human approval gates to agentic science workflows, with eight biological/chemical use-case demos.
T0 review reviewed 2026-08-01 challenge →
load-bearing objection An honest systems paper whose integrated design is real, but whose evidence trails are thinner than the word 'demonstrated' implies; §5.2's admission about incomplete provenance capture is the load-bearing gap. the 5 major comments →
SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
Core claim
SciForge's central claim is architectural: agents fail to support real research not because of model capability but because research state—artifacts, actions, decisions, and evidence traces—is not preserved as a coherent, auditable whole. The paper proposes a five-pillar workbench whose graphical interface is reserved for human judgment while search, parsing, model routing, execution, plotting, writing, and presentation generation run as modular agent-accessible services. The load-bearing device is the Evidence DAG, a claim-source-reasoning graph built automatically from agent turns, snapshotted into a goal-scoped Project DAG that merges findings across sessions and carries human review deci
What carries the argument
The machinery that carries the argument is the two-layer evidence graph plus the router. The Evidence DAG is a thread-scoped graph automatically compiled from completed agent turns, linking claims to source assertions, reasoning nodes, and support or contradiction edges, with node-level provenance visible to the researcher. The Project DAG consumes immutable Evidence DAG snapshots across sessions, merges equivalent findings while preserving independent source paths, and provides candidate and certified release gates tied to human decision records. The Scientific Model Router adds translate-then-reason: scientific files are detected by modality and passed through specialized translators that
Load-bearing premise
The whole audit value proposition rests on the Evidence DAG capturing complete, faithful provenance from agent runs; the paper concedes that software, parameters, environment, logs, and seed are not automatically populated, so the graph is only as auditable as whatever the runtime happens to record.
What would settle it
Run a controlled scripted session with known steps—specific tool versions, parameters, and a fixed random seed—then export the Evidence DAG and compare its provenance fields against ground truth; or inject an unsupported claim into an agent session and check whether the read-only audit flags it as ungrounded before any human review.
If this is right
- If the architecture holds, agent-generated scientific claims become inspectable end to end: each conclusion can be traced back to the files, scripts, model calls, parameters, and human approvals that produced it.
- Multi-day, multi-session research stays continuous: goals, evidence snapshots, and review decisions persist in the Project DAG, so work does not reset when a session ends.
- Scientific file formats stop being dead letters to language agents; proteins, structures, molecules, and single-cell data enter as structured expert observations rather than raw prompts, with unsupported formats failing closed.
- The same evidence-aware control chain can restructure routine research tasks—manuscript review and rebuttal, paper reproduction, figure and presentation generation—into auditable revision packages.
- Honest outcome reporting becomes a governance feature: in the molecular optimization use case the pre-registered primary criterion was not met, and that failure was recorded transparently rather than adjusted.
Where Pith is reading between the lines
- If provenance capture were made complete and enforced at execution time—software version, parameters, environment, logs, and seed—the audit sidecar would become a reproducibility linter for agent runs; the paper's own limitation note suggests this is the natural next step and a testable extension.
- The audit claim is only as strong as the capture layer; a stress test with deliberately planted fabricated claims or omitted provenance would show whether the DAG exposes them before a human reviews.
- The thin-GUI-plus-modular-services division of labour is not science-specific; the same pattern could govern other evidence-heavy, approval-gated domains such as regulated analysis or clinical decision support, though the paper does not say this.
- The eight use cases vary in maturity; adopting pre-registered success criteria uniformly would let the workbench generate its own evaluation evidence rather than relying on narrative demonstrations.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper describes SciForge, a local-first, AI-native research workbench designed to keep scientific artifacts, agent actions, and human decisions in one auditable state. Its architecture has five layers: thin human-judgment interaction surfaces; six research capability patterns; a core engine with an Agent Runtime, Workflow Engine, scoped memory, and evidence governance; a Scientific Model Router with translate-then-reason handling for protein, structure, molecule, and single-cell modalities; and local-first infrastructure. The claimed differentiators are goal-scoped decision governance, thread-level Evidence DAGs linked to a goal-scoped Project DAG, release gates, and multimodal scientific ingress. The system is 'demonstrated through eight end-to-end user cases,' including a multi-day agentic meiosis research sprint, protein contact prediction, reviewer/rebuttal support, MCFST paper reproduction, cross-scale cell atlas construction, de novo protein design, molecular optimization, and genome-to-BGC discovery. The paper is unusually candid: several use cases explicitly report unmet criteria, post-hoc thresholds, provenance mismatches, and the absence of baselines or third-party validation.
Significance. If the architecture works as claimed, SciForge would be a useful contribution to the emerging category of persistent, auditable AI research environments: it goes beyond single-session chat by linking claims to files, model calls, parameters, and human approvals through PROV-aligned Evidence/Project DAGs, and it makes a concrete open-source release with per-use-case repositories. The paper's main strengths are its modular architecture, fail-closed modality routing, explicit release-gate semantics, and the exceptionally transparent reporting of negative or ambiguous results (e.g., §4.7's unmet primary criterion, §4.4's post-hoc threshold, §4.6's provenance mismatch). These strengths are, however, self-assessed: there are no baselines, ablations, user studies, third-party reproductions, or capture-completeness benchmarks, and the central 'auditable traceability' claim depends on runtime provenance capture that the paper itself concedes is incomplete. The contribution is therefore best read as an architecture proposal with pilot illustrations, not as a validated demonstration of the auditability value proposition.
major comments (5)
- [§5.2, §3.4] The central claim of auditable traceability rests on automatic Evidence DAG capture, but §5.2 concedes that the DAG 'depends on runtime capture; software, parameters, environment, log/output, and seed are not automatically populated.' These are the same provenance fields the architecture (§3.4, §1.2) promises to attach to every agent action. Without a capture-completeness benchmark or independent audit measuring what fraction of model calls, tool invocations, file reads, and parameters actually appear in the DAG, the 'automatically constructs' claim is unsupported. I recommend either instrumenting the runtime to report capture coverage on the existing use cases or sharply narrowing the auditability claim in the abstract/contributions.
- [§4.4] The guided reproduction case is presented as an 'independently auditable evidence' demonstration, yet the reported evidence is internally inconsistent: the text says 25 independent training runs, verify.py reports n_runs=5, the 0.05 success threshold was applied post-hoc, and the selection rule for the five runs is not pre-specified. The best ARI of 0.7007 differs from the paper's 0.693 by +0.0077 with no stability analysis. This is precisely the kind of irreproducible verification record the workbench is supposed to prevent. The case should be reframed as a feasibility prototype (as the text later says) and the verification contract should be fixed with a pre-registered selection rule and reconciled run counts before the audit trail is presented as a positive demonstration.
- [§4.6] The self-audit reveals a provenance mismatch: the agent's final narrative cited ProteinMPNN scores (1.0363 and 1.0421) from unverified sequence samples rather than the scores of the sequences actually submitted to Boltz-2 (1.1039 and 1.1481). This demonstrates the failure mode in which an Evidence DAG can look coherent while pointing to the wrong artifacts. Since this is the flagship protein-design demonstration, the paper needs to show how the DAG and audit sidecar would surface such mismatches automatically — e.g., by linking each claimed score to the exact sequence hash and run record — rather than relying on a separate post-hoc audit.
- [§4.7] The abstract lists 'molecular optimization' as a demonstrated flagship scenario, but the primary pre-registered criterion was not met: the best improvement is −1.7 kcal/mol against a −2.0 threshold, and the text notes the erlotinib docking-score variance across replicates was ±2.0 kcal/mol. The observed improvement is therefore within the noise floor of the docking protocol. The honest reporting is commendable, but 'demonstrated' overstates the outcome; this should be described as a pilot/SAR illustration with the quantitative limitation stated in the abstract.
- [§5.2, §2.4] The paper's positioning against Claude Science, OmicOS, OmicsClaw, Operon, and research agents is based on a feature-level reading of public materials, and §5.2 explicitly states there are 'no baselines, ablation studies, user studies, or third-party reproductions.' For a systems paper whose contribution is the workbench claim, this leaves the main differentiators — evidence governance, capture completeness, and goal-scoped release semantics — unbenchmarked. I would not require a full user study for acceptance, but an evaluation section comparing at least provenance capture and task completion on a small standardized set (e.g., rerunning one case under two runtimes) would materially strengthen the central claim.
minor comments (6)
- [§4.1] The 'informal post-hoc literature mapping' is used to support the 23-gene atlas's plausibility. The paper already notes it was informal and not pre-registered; please mark it clearly as anecdotal or move it to supplementary so it is not read as validation.
- [§4.2] The repository URL 'https://github.com/BruthYU/autoresearch base' contains a space and is likely 'autoresearch-base'. Please fix the link.
- [§4.7] Figure 13 is referenced before Figure 12 in the text ('scaffold modification strategy is depicted in Fig. 13' precedes 'Figure 12 summarizes...'). Renumber or adjust the in-text references.
- [§3.2, §4] The abbreviation 'PI' is used without definition on first use; please define it (e.g., principal investigator) in §3 or §4.
- [Title page, Appendix C] The title-page attribution 'Written by SciForge with DeepSeek-v4-pro (text) and gpt-image-2 (figures), Guided and Verified by Humans' appears to conflict with Appendix C, which lists human manuscript contributors. Clarify the drafting/verification workflow or move the model attribution to a footnote.
- [Table 1] The dagger marker for the planned 'Team Workspace' capability is ambiguous about which column it annotates. Move the dagger into the SciForge column and state in the caption that the capability is planned, not implemented.
Circularity Check
No circular derivation; the workbench's core auditability claim rests on a stated validation gap, not on circular reasoning.
full rationale
SciForge is a systems paper: it does not derive predictions from fitted parameters or mathematical inputs. The central claim is architectural—that an Evidence DAG/Project DAG can make agent workflows auditable—and it is demonstrated through case studies. The load-bearing assumption is that runtime provenance capture is complete and faithful. The paper explicitly concedes in §5.2 that 'the Evidence DAG schema can represent complete provenance but depends on runtime capture; software, parameters, environment, log/output, and seed are not automatically populated,' and that 'no baselines, ablation studies, user studies, or third-party reproductions have been conducted.' These are validation gaps, not circular reductions. The MCFST reproduction (§4.4) discloses that the 0.05 success threshold was applied post-hoc and that verify.py reports n_runs=5, conflicting with 25 total predictions; this is an admitted statistical weakness, not a hidden fit called a prediction. The protein-design audit (§4.6) discloses a provenance mismatch between cited and executed ProteinMPNN scores; this is a self-audit finding that lowers confidence but does not define a result in terms of its own inputs. No self-citation chain or uniqueness theorem is load-bearing; external references are standard tools/translators (RFdiffusion, ProteinMPNN, Boltz-2, antiSMASH, etc.), and the workbench code is open-source. Therefore there is no significant circularity.
Axiom & Free-Parameter Ledger
free parameters (2)
- MCFST success threshold =
0.05
- BGC rubric weights/thresholds (v0.1) =
unknown
axioms (3)
- domain assumption The four domain translators (Esm2Text, Prot2Text, BioT5+, C2S) produce structured expert observations sufficiently reliable for agent reasoning about scientific files.
- domain assumption Automatically captured Evidence Snapshots and the Evidence DAG accurately represent the provenance of agent actions (software, parameters, environment, logs, seed).
- domain assumption External scientific resources (UniProt, Reactome, DepMap, antiSMASH, MIBiG, BiG-SCAPE, PDB, structure-prediction models) are treated as ground truth for use-case conclusions.
invented entities (2)
-
Evidence DAG (thread-scoped)
no independent evidence
-
Project DAG (goal-scoped)
no independent evidence
Cite this review
Pith. "Pith review of SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery." pith.science (2026). https://pith.science/paper/ME5ZDBA2
@misc{pith2026260716038,
author = {Pith},
title = {Pith review of: SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery},
year = {2026},
howpublished = {\url{https://pith.science/paper/ME5ZDBA2}},
note = {Machine review of arXiv:2607.16038}
}
read the original abstract
Scientific work increasingly spans heterogeneous artifacts -- papers, code, datasets, scientific file formats, model outputs, figures, manuscripts, and team decisions -- yet general-purpose AI assistants rarely preserve these objects as a coherent, auditable research state. We present SciForge, a multimodal research-native AI workbench that reserves the graphical interface for human judgment while search, parsing, model routing, workflow execution, plotting, writing, and presentation generation run as modular agent-accessible services. SciForge is built around five pillars: (i) \emph{goal-scoped scientific decision governance} for \textbf{goal-oriented} research, with review gates and shared review surfaces; (ii) \emph{translate-then-reason} for \textbf{multimodal} input, routing scientific objects through domain translators before the agent reasons; (iii) \emph{evidence governance} for \textbf{auditable} traceability, linking claims to provenance chains and audit findings; (iv) \emph{collaborative team science} for \textbf{collaborative} research, enabling multi-role decision governance, with shared team workspaces planned for future releases; and (v) \emph{real-world application scenarios} for \textbf{practical} impact, demonstrated through eight end-to-end user cases, with flagship demonstrations including multi-day agentic research sprints for gene discovery, AI-guided de novo protein design, molecular optimization, and genome-to-BGC discovery. The system combines a thin interaction layer, contextual research capability patterns, an Agent Runtime and Workflow Engine, an Evidence-DAG audit sidecar and a Scientific Model Router. SciForge currently runs as a desktop application, with mobile supervision support; future releases will deepen team collaboration. The system is open-source and available at https://github.com/AGI4Sci/SciForge
Reference graph
Works this paper leans on
-
[1]
arXiv preprint arXiv:2307.14367
Abdine H, Chatzianastasis M, Bouyioukos C, et al (2023) Prot2Text: Multi- modal protein’s function generation with GNNs and transformers. arXiv preprint arXiv:2307.14367
Pith/arXiv arXiv 2023
-
[2]
Abramson J, Adler J, Dunger J, et al (2024) Accurate Structure Prediction of Biomolecular Interactions with AlphaFold 3. Nature 630:493–500. https://doi.or g/10.1038/s41586-024-07487-w
-
[3]
Bioinformatics 31(13):2214–2216
Alhossary A, Handoko SD, Mu Y, et al (2015) Fast, Accurate, and Reliable Molecular Docking with QuickVina 2. Bioinformatics 31(13):2214–2216. https: 36 //doi.org/10.1093/bioinformatics/btv082
-
[4]
Available at https://allenai.org/bl og/ai2-scholarqa (accessed 4 July 2026)
Allen Institute for AI (2025) Ai2 ScholarQA. Available at https://allenai.org/bl og/ai2-scholarqa (accessed 4 July 2026)
2025
-
[5]
Available at https: //www.anthropic.com/news/model-context-protocol (accessed 4 July 2026)
Anthropic (2024) Introducing the Model Context Protocol. Available at https: //www.anthropic.com/news/model-context-protocol (accessed 4 July 2026)
2024
-
[6]
Available at https://www.anthropic.com/news/claude-science-ai-workbench (accessed 4 July 2026)
Anthropic (2026) Claude Science: An AI Workbench for Scientists. Available at https://www.anthropic.com/news/claude-science-ai-workbench (accessed 4 July 2026)
2026
-
[7]
Asai A, et al (2026) Synthesizing Scientific Literature with Retrieval-Augmented Language Models. Nature 650(8103):857–863. https://doi.org/10.1038/s41586-025 -10072-4
-
[8]
Journal of Medicinal Chemistry 53(7):2719–2740
Baell JB, Holloway GA (2010) New Substructure Filters for Removal of Pan Assay Interference Compounds (PAINS) from Screening Libraries and for Their Exclusion in Bioassays. Journal of Medicinal Chemistry 53(7):2719–2740. https: //doi.org/10.1021/jm901137j
-
[9]
Bickerton GR, Paolini GV, Besnard J, et al (2012) Quantifying the Chemical Beauty of Drugs. Nature Chemistry 4:90–98. https://doi.org/10.1038/nchem.1243
-
[10]
Nucleic Acids Research 51(W1):W46–W50
Blin K, Shaw S, Augustijn HE, et al (2023) antiSMASH 7.0: New and Improved Predictions for Detection, Regulation, Chemical Structures and Visualisation. Nucleic Acids Research 51(W1):W46–W50. https://doi.org/10.1093/nar/gkad344
-
[11]
Boiko DA, MacKnight R, Kline B, et al (2023) Autonomous Chemical Research with Large Language Models. Nature 624:570–578. https://doi.org/10.1038/s415 86-023-06792-0
doi:10.1038/s415 2023
-
[12]
Nature Machine Intelligence 6(5):525–535
Bran AM, Cox S, Schilter O, et al (2024) Augmenting Large Language Models with Chemistry Tools. Nature Machine Intelligence 6(5):525–535. https://doi.org/ 10.1038/s42256-024-00832-8
-
[13]
Chen Z, Chen S, Ning Y, et al (2025) ScienceAgentBench: Toward Rigorous Assess- ment of Language Agents for Data-Driven Scientific Discovery. In: International Conference on Learning Representations, available at https://proceedings.iclr.cc/ paper files/paper/2025/hash/f12b4df26344f3be803c06b555252efe-Abstract-Con ference.html
2025
-
[14]
Available at https://github.com/delibae/claude-prism (accessed 4 July 2026) 37
ClaudePrism Contributors (2026) ClaudePrism: Desktop Workspace for Scientific Writing. Available at https://github.com/delibae/claude-prism (accessed 4 July 2026) 37
2026
-
[15]
Available at https://help.consensus .app/en/articles/12641232-research-agent (accessed 4 July 2026)
Consensus (2026) Consensus Research Agent. Available at https://help.consensus .app/en/articles/12641232-research-agent (accessed 4 July 2026)
arXiv 2026
-
[16]
Dauparas J, Anishchenko I, Bennett N, et al (2022) Robust Deep Learning- Based Protein Sequence Design Using ProteinMPNN. Science 378(6615):49–56. https://doi.org/10.1126/science.add2187
-
[17]
Available at https://elicit.com/ (accessed 4 July 2026)
Elicit (2026) Elicit Research Agent. Available at https://elicit.com/ (accessed 4 July 2026)
2026
-
[18]
Journal of Cheminformatics 1:8
Ertl P, Schuffenhauer A (2009) Estimation of Synthetic Accessibility Score of Drug-Like Molecules Based on Molecular Complexity and Fragment Contributions. Journal of Cheminformatics 1:8. https://doi.org/10.1186/1758-2946-1-8
-
[19]
Available at https://www.futurehouse.org/news/launching-futurehouse-platform-ai-agents (accessed 4 July 2026)
FutureHouse (2025) Launching FutureHouse Platform AI Agents. Available at https://www.futurehouse.org/news/launching-futurehouse-platform-ai-agents (accessed 4 July 2026)
2025
-
[20]
Advanced Materials https://doi.org/10.1002/adma.202413523
Ghafarollahi A, Buehler MJ (2025) SciAgents: Automating Scientific Discovery through Bioinspired Multi-Agent Intelligent Graph Reasoning. Advanced Materials https://doi.org/10.1002/adma.202413523
-
[22]
Gillespie M, Jassal B, Stephan R, et al (2024) Reactome Pathway Knowledgebase
2024
-
[23]
Gottweis J, Weng WH, Daryin A, et al (2026) Accelerating Scientific Discovery with Co-Scientist. Nature 655(8122):487–496. https://doi.org/10.1038/s41586-026 -10644-y
-
[24]
Hayes T, Rao R, Akin H, et al (2025) Simulating 500 Million Years of Evolution with a Language Model. Science 387(6736):850–858. https://doi.org/10.1126/scie nce.ads0018
doi:10.1126/scie 2025
-
[25]
Nature Methods 18(11):1342–1351
Hu J, Li X, Coleman K, et al (2021) SpaGCN: Integrating Gene Expression, Spatial Location and Histology to Identify Spatial Domains and Spatially Variable Genes by Graph Convolutional Network. Nature Methods 18(11):1342–1351. https: //doi.org/10.1038/s41592-021-01255-8
-
[26]
bioRxiv https://doi.org/10.1101/2025.05.30.656746 38
Huang K, Zhang S, Wang H, et al (2025) Biomni: A General-Purpose Biomedical AI Agent. bioRxiv https://doi.org/10.1101/2025.05.30.656746 38
-
[27]
W3c member submission, World Wide Web Consortium, available at https://www.w3.org/submissions/201 3/SUBM-prov-json-20130424/
Huynh TD, Jewell MO, Keshavarz AS, et al (2013) The PROV-JSON Serialization: A JSON Representation for the PROV Data Model. W3c member submission, World Wide Web Consortium, available at https://www.w3.org/submissions/201 3/SUBM-prov-json-20130424/
2013
-
[28]
Ifargan T, Hafner L, Kern M, et al (2025) Autonomous LLM-Driven Research from Data to Human-Verifiable Research Papers. NEJM AI 2(1). https://doi.org/ 10.1056/AIoa2400555
-
[29]
Jansen P, Cˆ ot´ e MA, Khot T, et al (2024) DiscoveryWorld: A Virtual Environ- ment for Developing and Evaluating Automated Scientific Discovery Agents. In: Advances in Neural Information Processing Systems, pp 10088–10116, https: //doi.org/10.52202/079017-0324
-
[30]
Nature 596(7873):583–589
Jumper J, Evans R, Pritzel A, et al (2021) Highly Accurate Protein Structure Prediction with AlphaFold. Nature 596(7873):583–589. https://doi.org/10.1038/ s41586-021-03819-2
2021
-
[31]
Nature Reviews Microbiology 17:167–180
Keller NP (2019) Fungal Secondary Metabolism: Regulation, Function and Drug Discovery. Nature Reviews Microbiology 17:167–180. https://doi.org/10.1038/s4 1579-018-0121-1
doi:10.1038/s4 2019
-
[32]
Preprint at https://arxiv.org/abs/ 2409.13740
L´ ala J, O’Donoghue O, Shtedritski A, et al (2024) Language Agents Achieve Superhuman Synthesis of Scientific Knowledge. Preprint at https://arxiv.org/abs/ 2409.13740
Pith/arXiv arXiv 2024
-
[33]
https://ww w.rdkit.org; software archive https://doi.org/10.5281/zenodo.591637
Landrum G (2016) RDKit: Open-Source Cheminformatics Software. https://ww w.rdkit.org; software archive https://doi.org/10.5281/zenodo.591637
-
[34]
Lin Z, Akin H, Rao R, et al (2023) Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379(6637):1123–1130. https: //doi.org/10.1126/science.ade2574
-
[35]
Advanced Drug Delivery Reviews 46(1–3):3–26
Lipinski CA, Lombardo F, Dominy BW, et al (2001) Experimental and Compu- tational Approaches to Estimate Solubility and Permeability in Drug Discovery and Development Settings. Advanced Drug Delivery Reviews 46(1–3):3–26. https://doi.org/10.1016/S0169-409X(00)00129-0
-
[36]
Nature Communications 14(1):1155
Long Y, Ang KS, Li M, et al (2023) Spatially Informed Clustering, Integration, and Deconvolution of Spatial Transcriptomics with GraphST. Nature Communications 14(1):1155. https://doi.org/10.1038/s41467-023-36796-3
-
[37]
Lu C, Lu C, Lange RT, et al (2026) Towards End-to-End Automation of AI Research. Nature 651:914–919. https://doi.org/10.1038/s41586-026-10265-5
-
[38]
W3C Recommendation, https://www.w3.org/TR/prov-dm/ 39
Moreau L, Missier P (2013) PROV-DM: The PROV Data Model. W3C Recommendation, https://www.w3.org/TR/prov-dm/ 39
2013
-
[39]
Nature Chemical Biology 16:60–68
Navarro-Mu˜ noz JC, Selem-Mojica N, Mullowney MW, et al (2020) A Computa- tional Framework to Explore Large-Scale Biosynthetic Diversity. Nature Chemical Biology 16:60–68. https://doi.org/10.1038/s41589-019-0400-9
-
[40]
Available at https://docs.omicos.cn/zh/part 1/01-overview.html (accessed 4 July 2026)
OmicOS (2026) OmicOS Overview. Available at https://docs.omicos.cn/zh/part 1/01-overview.html (accessed 4 July 2026)
2026
-
[41]
Available at https://docs.omicos.cn/zh/pa rt1/08-starting-cli.html (accessed 4 July 2026)
OmicOS (2026) Starting OmicOS CLI. Available at https://docs.omicos.cn/zh/pa rt1/08-starting-cli.html (accessed 4 July 2026)
2026
-
[42]
Available at https://docs.omicos.cn/zh/ part1/07-starting-serve.html (accessed 4 July 2026)
OmicOS (2026) Starting OmicOS Serve. Available at https://docs.omicos.cn/zh/ part1/07-starting-serve.html (accessed 4 July 2026)
2026
-
[43]
Available at https://github.com/TianGzlab/OmicsClaw (accessed 4 July 2026)
OmicsClaw Contributors (2026) OmicsClaw: Local-First Multi-Omics Research Assistant. Available at https://github.com/TianGzlab/OmicsClaw (accessed 4 July 2026)
2026
-
[44]
Papalexi E, Mimitou EP, Butler A W, et al (2021) Characterizing the Molecular Regulation of Inhibitory Immune Checkpoints with Multimodal Single-Cell Screens. Nature Genetics 53:322–331. https://doi.org/10.1038/s41588-021-00778-2
-
[45]
Biochemical Journal 448(3):417–423
Park JH, Liu Y, Lemmon MA, et al (2012) Erlotinib Binds Both Inactive and Active Conformations of the EGFR Tyrosine Kinase Domain. Biochemical Journal 448(3):417–423. https://doi.org/10.1042/BJ20121513
-
[46]
bioRxiv https://doi.org/10.1101/2025.06.14 .659707
Passaro S, Corso G, Wohlwend J, et al (2025) Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction. bioRxiv https://doi.org/10.1101/2025.06.14 .659707
-
[47]
arXiv preprint arXiv:2402.17810
Pei Q, Wu L, Gao K, et al (2024) BioT5+: Towards generalized biological understanding with IUPAC integration and multi-task tuning. arXiv preprint arXiv:2402.17810
Pith/arXiv arXiv 2024
-
[48]
Nature Biomedical Engineering 10:245–258
Qu Y, Huang K, Yin M, et al (2026) CRISPR-GPT for Agentic Automation of Gene-Editing Experiments. Nature Biomedical Engineering 10:245–258. https: //doi.org/10.1038/s41551-025-01463-z
-
[49]
bioRxiv https://doi.org/10.1101/2025.04.14.64885 0
Rizvi SA, Levine D, Patel A, et al (2025) Scaling large language models for next- generation single-cell analysis. bioRxiv https://doi.org/10.1101/2025.04.14.64885 0
-
[50]
Rocklin GJ, Chidyausiku TM, Goreshnik I, et al (2017) Global Analysis of Protein Folding Using Massively Parallel Design, Synthesis, and Testing. Science 357(6347):168–175. https://doi.org/10.1126/science.aan0693
-
[51]
Journal of Chemi- cal Information and Modeling 50(5):742–754
Rogers D, Hahn M (2010) Extended-Connectivity Fingerprints. Journal of Chemi- cal Information and Modeling 50(5):742–754. https://doi.org/10.1021/ci100050 t 40
-
[52]
PLoS Computational Biology 9(10):e1003285
Sandve GK, Nekrutenko A, Taylor J, et al (2013) Ten Simple Rules for Repro- ducible Computational Research. PLoS Computational Biology 9(10):e1003285. https://doi.org/10.1371/journal.pcbi.1003285
-
[53]
In: Findings of the Association for Computational Linguistics: EMNLP 2025
Schmidgall S, Su Y, Wang Z, et al (2025) Agent Laboratory: Using LLM Agents as Research Assistants. In: Findings of the Association for Computational Linguistics: EMNLP 2025. Association for Computational Linguistics, pp 5977–6043, https: //doi.org/10.18653/v1/2025.findings-emnlp.320
-
[54]
St ˚ ahl PL, Salm´ en F, Vickovic S, et al (2016) Visualization and Analysis of Gene Expression in Tissue Sections by Spatial Transcriptomics. Science 353(6294):78–82. https://doi.org/10.1126/science.aaf2403
-
[55]
Nature 646:716–723
Swanson K, Wu W, Bulaong NL, et al (2025) The Virtual Lab of AI Agents Designs New SARS-CoV-2 Nanobodies. Nature 646:716–723. https://doi.org/10.1 038/s41586-025-09442-9
2025
-
[56]
Available at https://github.com/swaruplab/operon (accessed 4 July 2026)
Swarup Lab (2026) Operon: AI-Powered IDE for Bioinformatics. Available at https://github.com/swaruplab/operon (accessed 4 July 2026)
2026
-
[57]
Szymanski NJ, Rendy B, Fei Y, et al (2023) An Autonomous Laboratory for the Accelerated Synthesis of Inorganic Materials. Nature 624:86–91. https://doi.org/ 10.1038/s41586-023-06734-w
-
[58]
Tsherniak A, Vazquez F, Montgomery PG, et al (2017) Defining a Cancer Depen- dency Map. Cell 170(3):564–576.e16. https://doi.org/10.1016/j.cell.2017.06.0 10
-
[59]
UniProt Consortium (2025) UniProt: The Universal Protein Knowledgebase in
2025
-
[60]
Watson JL, Juergens D, Bennett NR, et al (2023) De Novo Design of Protein Structure and Function with RFdiffusion. Nature 620:1089–1100. https://doi.org/ 10.1038/s41586-023-06415-8
-
[61]
Wilkinson MD, Dumontier M, Aalbersberg IJ, et al (2016) The F AIR Guiding Prin- ciples for Scientific Data Management and Stewardship. Scientific Data 3:160018. https://doi.org/10.1038/sdata.2016.18
-
[62]
Xu H, Fu H, Long Y, et al (2024) Unsupervised Spatially Embedded Deep Representation of Spatial Transcriptomics. Genome Medicine 16(1):12. https: //doi.org/10.1186/s13073-024-01283-x
-
[63]
Preprint at https://arxi v.org/abs/2504.08066 41
Yamada Y, Lange RT, Lu C, et al (2025) The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search. Preprint at https://arxi v.org/abs/2504.08066 41
Pith/arXiv arXiv 2025
-
[64]
arXiv preprint ArXiv:2604.27646v1
Yang K, et al (2026) Benchmarking virtual cell models for in-the-wild perturbation response. arXiv preprint ArXiv:2604.27646v1
Pith/arXiv arXiv 2026
-
[65]
Zdouc MM, Blin K, Louwen NLL, et al (2025) MIBiG 4.0: Advancing Biosynthetic Gene Cluster Curation through Global Collaboration. Nucleic Acids Research 53(D1). https://doi.org/10.1093/nar/gkae1115
-
[66]
Zhang Z, Duan H, Gao X (2026) MCFST: Spatial domain identification method based on multi-view graph convolutional network and graph fusion network. Bioinformatics 42(7). https://doi.org/10.1093/bioinformatics/btag469
-
[67]
Nature Biotechnology 39(11):1375–1384
Zhao E, Stone MR, Ren X, et al (2021) Spatial Transcriptomics at Subspot Resolution with BayesSpace. Nature Biotechnology 39(11):1375–1384. https: //doi.org/10.1038/s41587-021-00935-2 42
-
[2025]
https://doi.org/10.1093/nar/gk ae1010
Nucleic Acids Research 53(D1):D609–D617. https://doi.org/10.1093/nar/gk ae1010
This paper was first reviewed by deepseek-v4-flash on August 1, 2026.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.