Pith. sign in

REVIEW 3 major objections 6 minor 57 references

Automating Exploratory Multiomics Research via Language Models

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read An LLM-orchestrated pipeline can carry out open-ended multiomics exploration and produce hypotheses that are verifiable in external cancer cohorts.

desk verdict Promising architecture for open-ended multiomics hypothesis generation, but the external validation is partly in-sample for BRCA/LSCC and the evaluation needs human calibration. read the letter →

arxiv 2506.07591 v1 pith:TOWQDR35 submitted 2025-06-09 cs.AI q-bio.QM

Shang Qu , Ning Ding , Linhai Xie , Yifei Li , Zaoqu Liu , Kaiyan Zhang , Yibai Xiong , Yuxin Zuo
show 8 more authors
Zhangren Chen Ermo Hua Xingtai Lv Youbang Sun Yang Li Dong Li Fuchu He Bowen Zhou
This is my paper
classification cs.AIq-bio.QM
keywords largelanguagemodelsmultiomicshypothesisgenerationproteogenomicsautonomousresearchCPTACvalidationscientificdiscovery
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that PROTEUS, a fully automated system built around a large language model, can take raw clinical multiomics files and, with no human-set research question, produce concrete scientific hypotheses. Applied to 10 cancer datasets, it generated 360 hypotheses, and the authors argue that these balance statistical reliability with novelty: external CPTAC cohorts corroborated many claims, and automatic literature-aware scoring rated the hypotheses highly. If this holds, the bottleneck in multiomics discovery shifts from data analysis to choosing which machine-generated leads to test. The key caveat is that the system's own self-scoring is a central part of the quality evidence.

What carries the argument

The central mechanism is a pair of graph structures that organize the entire research process: a static relationship graph listing biological entity types and the allowed relationship types among them, and a dynamic conclusion graph recording the specific entities and relationships the system has actually analyzed. Five LLM-driven modules — Explorer, Hypothesizer, Decomposer, Validator, and Integrator — use these graphs to iterate, with the conclusion graph serving as long-term memory so the system can deepen or diversify its investigations instead of repeating past analyses.

What would settle it

Take 50 of the 360 hypotheses at random, have clinical biologists who were not involved rate them on the same five scales, and compare the two score distributions; if the human scores trail GPT-4o's by a wide margin on novelty or significance, the claimed balance is an artifact of self-assessment.

Watch

Extended reading notes

Core claim

PROTEUS is presented as the first fully automated end-to-end system for exploratory clinical proteogenomics research: it ingests raw protein, transcript, phosphosite, variant, and clinical-feature tables; an LLM (gpt-4o) orchestrates five modules that choose a general research direction, narrow it to a specific hypothesis, decompose it into single relationships, validate each relationship by calling one of 41 bioinformatics tools with retry logic, and integrate results into final hypotheses. The paper reports that across 10 datasets PROTEUS produced 360 hypotheses, that on 5 datasets with matching CPTAC cohorts a majority of validation results supported the hypotheses (over 70% support or weak support on four of the five), and that on all five automatic scoring metrics (literature alignment, logical coherence, scientific novelty, biological significance, general quality) PROTEUS surpassed an LLM baseline that writes its own Python analysis code. The authors therefore claim PROTEUS produces hypotheses that are both statistically grounded and open-endedly novel.

Load-bearing premise

The evaluation's credibility depends on GPT-4o scoring hypotheses that it generated itself, using only PubMed abstracts, on subjective scales like novelty and significance, without any human or independent calibration.

Editorial extensions

If this is right

  • If the claim holds, multiomics studies that currently require weeks of manual bioinformatics work could be triaged automatically, with only the most promising machine-generated hypotheses forwarded to human experts.
  • The graph-structured pipeline should transfer to other data-rich scientific domains, such as chemistry, spatial biology, or epidemiology, where the research question itself is what needs to be discovered.
  • Because every validated relationship is recorded in the conclusion graph, the system's output can serve as a machine-readable audit trail showing exactly which statistical edges support each hypothesis.
  • The finding that increased hypothesis complexity did not destroy logical coherence suggests LLM orchestrators can reliably chain many tool calls, a result that matters for any multi-step autonomous system.
  • If adopted as a routine tool, PROTEUS could let researchers explore far more research directions per dataset than is feasible manually, shifting the practical limit on discovery from analyst effort to the supply of independent validation cohorts.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct test not in the paper: rerun PROTEUS on the same 10 datasets with different random seeds or a different base LLM and measure overlap between the resulting hypothesis sets; high overlap would suggest the discoveries are data-driven rather than model-idiosyncratic.
  • The paper's own evaluation leaves the self-scoring issue open: the same model that generates a hypothesis also judges its novelty and significance. Having independent human experts rate a random subset on the same scales would reveal whether the claimed balance is genuine or an artifact of self-assessment.
  • If matching external cohorts ever become available for the five datasets that currently lack them (SCLC, HCC, CCA, iCCA, mCRC), the distribution of support statuses should resemble that seen on ccRCC, LSCC, and LUAD; a substantially worse result on those would indicate that success depends on dataset recency or cohort quality rather than on the system itself.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper introduces PROTEUS, a fully automatic LLM-orchestrated system that takes raw clinical multiomics data files as input and produces open-ended, data-driven scientific hypotheses. PROTEUS structures the research process with five modules (Explorer, Hypothesizer, Decomposer, Validator, Integrator) and uses a static relationship graph and a dynamic conclusion graph to organize biological entities, tools, and findings. The authors apply PROTEUS to 10 clinical multiomics datasets, generating 360 hypotheses, and evaluate them in two ways: validation against CPTAC cohorts for five cancer types, and automatic LLM scoring along five metrics (Literature Alignment, Logical Coherence, Scientific Novelty, Biological Significance, General Quality). The central claim is that PROTEUS produces hypotheses that balance statistical reliability and novelty, outperforming two LLM-centered baseline settings on all five scoring metrics while maintaining comparable or better external validation support rates.

Significance. If the central claims hold, PROTEUS would be a meaningful step toward autonomous scientific discovery in multiomics: the system is fully automated, integrates a nontrivial suite of bioinformatics tools, uses graph-structured memory to manage complex multi-step analyses, and produces a large corpus of specific, testable hypotheses. The paper also contributes a scalable, quantitative evaluation pipeline that combines external cohort statistics with literature-grounded LLM scoring, and it includes ablations across base models and tool-retry behavior. These are valuable features for the emerging area of LLM-driven scientific agents. However, the current evidence does not fully establish the claimed balance between reliability and novelty because of the external-validity and evaluation-bias issues detailed below. With these addressed, the work could be an important reference for autonomous multiomics analysis.

major comments (3)
  1. [2.2 / Table 1 / Section 4.4] The claim that PROTEUS hypotheses are 'verifiable in external cohorts' (Section 2.2) is not established for at least two of the five evaluated datasets. The BRCA discovery dataset is Mertins et al. [29] and the LSCC discovery dataset is Stewart et al. [33]; both are CPTAC studies, and the CPTAC validation data in Section 4.4 is retrieved from the `cptac` package for the same cancer types. The manuscript reports no sample-overlap analysis, no exclusion of overlapping patients, and no data/code release that would allow such a check. Since Section 2.2's support rates for BRCA and LSCC (and the 'balance reliability and novelty' claim) rely on this validation being external, the authors should either demonstrate non-overlap of samples or rerun the validation on truly independent cohorts.
  2. [2.3 / Section 4.5] The automatic hypothesis scoring uses GPT-4o to evaluate hypotheses that GPT-4o itself generated (Section 2.3 and Section 4.5). No human validation or independently scored subset is provided to calibrate these scores. Because the same model judges its own outputs along subjective dimensions such as Scientific Novelty and Biological Significance, the reported advantage over baselines may reflect systematic self-preference rather than true quality. I recommend a blinded human evaluation of a random subset (e.g., 50–100 hypotheses) or the use of a second, independent LLM, with inter-rater agreement reported.
  3. [2.2–2.3, Figures 2–3] The comparisons between PROTEUS and baselines, and across datasets, are reported without confidence intervals, effect sizes, or significance tests. For example, the support-rate percentages in Figure 2b and the average scores in Figure 3b are point estimates with no error bars; the claim that PROTEUS 'surpassed both baseline settings on all 5 metrics' (Section 2.3) is therefore not supported by statistical inference. Given the small number of hypotheses per dataset (36) and the shared analysis pipeline, these differences could be within sampling variation. Please add bootstrapped confidence intervals and appropriate paired tests to substantiate the comparative claims.
minor comments (6)
  1. [Section 4.2 vs. Section 2.1 / Table 2] Section 4.2 states 'a total of (36) tools' and '18 biological relationship types', while Section 2.1 and Table 2 state '41 bioinformatics tools' and '22 relationship types'; this numeric inconsistency should be resolved for reproducibility.
  2. [Section 2.2] The text repeatedly refers to 'LCSS' in the description of datasets with solid support rates; this should be corrected to 'LSCC'.
  3. [Abstract and Section 1] The system is described as fully automated, but Appendix A notes that data descriptions were written by humans; the manuscript should clarify what 'fully automated' excludes, since the input includes these minimal textual descriptions.
  4. [Section 4.4] The CPTAC data sources are identified only by names such as 'bcm', 'umich', and 'mssm'; the paper should specify the exact data versions or release identifiers to make the evaluation reproducible and to support the sample-overlap check requested above.
  5. [Figure 4] The figure contains duplicate subfigure label '(g)' and a stray '(g)' in the caption; the labeling should be corrected.
  6. [Discussion] The Discussion notes that predefined tools are the only avenue for data interaction, but the baseline uses free-form Python code generation; a short discussion of why PROTEUS does not also generate code would help contextualize this design choice.

Circularity Check

2 steps flagged · score 6.0 of 10

The novelty half of the central claim is scored by the same GPT-4o model that generated the hypotheses, and the 'external' CPTAC validation arguably reuses the discovery cohorts for BRCA and LSCC.

  1. other [Section 2.3 (Evaluation Method); Section 4.5; Section 2.1]
    "We used GPT-4o to conduct automatic hypothesis scoring according to 5 distinct metrics. ... In all main experiments, we use gpt-4o as the base LLM of PROTEUS."

    The evaluator and the generator are the same model. GPT-4o both produces PROTEUS's hypotheses and assigns the Literature Alignment, Logical Coherence, Scientific Novelty, Biological Significance, and General Quality scores, with only PubMed abstracts as reference. No independent rater, human or model, calibrates these scores. The 'novelty' and 'significance' scores are therefore the generator's self-report about its own outputs, not an externally grounded measurement; the claim that PROTEUS 'produces reliable and novel open-ended hypotheses' is supported on this axis by construction rather than by independent evidence.

  2. fitted input called prediction [Section 2.2 (Evaluation Method); Table 1; Section 4.4]
    "We mirror this method and use clinical cohort data from CPTAC to enable statistics-based verification of each individual hypothesis generated by PROTEUS. ... Proteogenomic landscape of squamous cell lung cancer [33] LSCC Yes 109 ... We used the cptac package to access and analyze datasets in CPTAC. For each of the 5 datasets (BRCA, ccRCC, GBM, LSCC, LUAD) involved..."

    For LSCC and BRCA, the discovery datasets fed to PROTEUS are themselves CPTAC studies: Stewart et al. [33] for LSCC and Mertins et al. [29] for BRCA. The 'external' validation then pulls CPTAC data from the cptac package for those same cancer types, and the paper reports no sample-ID overlap analysis or exclusion of overlapping samples. The support rates for these two datasets can therefore reflect in-sample agreement between hypotheses and the very cohorts they were generated from, so the heading 'PROTEUS produces hypotheses that are verifiable in external cohorts' is not established for two of the five datasets.

full rationale

The paper's engineering contribution is real and not circular: PROTEUS is a substantial LLM-orchestrated pipeline, and the CPTAC statistical validation for ccRCC, GBM, and LUAD is genuinely external to the discovery cohorts, providing independent statistical grounding for a large fraction of the generated hypotheses. However, the central characterization 'balance reliability and novelty' rests on two evaluation pillars, and both are partially circular. The LLM-scoring pillar uses GPT-4o to grade hypotheses that GPT-4o itself generated, with no independent calibration, so the novelty/significance scores are self-referential. The CPTAC pillar is external only if the validation cohorts do not overlap the discovery cohorts, but for LSCC (and likely BRCA) the discovery data are themselves CPTAC studies from the same sources the cptac package accesses, and no overlap check is reported. These are not cases of the derivation reducing to a fitted parameter by an explicit equation, but they are cases where the claimed empirical support reduces to the system's own inputs and outputs. Score 6 reflects partial circularity in the evaluation of the central claim rather than circularity in the system's internal derivation.

Assumptions & free parameters 2 free parameters · 4 assumptions · 2 invented entities

The paper introduces no new physical or mathematical entities, but it does introduce two software structures (relationship graph and conclusion graph) whose value is assumed. The main free parameters are statistical thresholds that the LLM adjusts during retries, which are fitted to the data in an outcome-dependent way. The axioms are domain assumptions about representation, LLM reliability, external validation, and LLM-based scoring.

free parameters (2)
  • Significance thresholds (max_p_value, max_adj_p_value, min_logFC, min_correlation) = Adjusted dynamically by the LLM during retries; no fixed value reported
    The Validator retry prompt instructs the model to adjust parameters when no successful results are found, and Section 2.4 mentions 'optimize any thresholds for result significance.' This creates an outcome-dependent threshold search that can inflate significant findings.
  • Survival expression-level cutoff = Selected by searching for the most significant log-rank or Cox p-value
    Section 4.2 (Survival Analysis) states that the tool experiments with different thresholds for classifying expression as high or low and selects the threshold yielding the most significant results. This is a per-test free parameter chosen by outcome.
assumptions (4)
  • domain assumption Relationships between biological entities (e.g., protein-clinic, variant-protein) are sufficient to represent multiomics research directions and hypotheses.
    The entire graph-based framework rests on this representational claim, introduced in Section 2.1 under 'Relationship and conclusion graphs.'
  • domain assumption gpt-4o possesses sufficient biological knowledge and tool-use reliability to plan, execute, and interpret bioinformatics analyses correctly.
    All five PROTEUS modules use gpt-4o as the base LLM (Section 2.1), and there is no independent verification of its tool outputs beyond the retry loop.
  • domain assumption CPTAC cohorts for the same cancer type constitute valid external validation data.
    Section 2.2 uses CPTAC cohorts to validate hypotheses. For some input datasets, especially BRCA (ref [29]), the input cohort may itself be CPTAC-derived, which would weaken the 'external' nature of the validation.
  • ad hoc to paper LLM scoring with PubMed abstracts is a valid proxy for hypothesis novelty and biological significance.
    Section 2.3 and Section 4.5 define the five scoring metrics and use gpt-4o as scorer, but no human expert calibration is provided.
invented entities (2)
  • Relationship graph (static graph of entity types and supported relationship types)
    purpose: Scaffolds research direction selection and tool filtering across all modules.
    A system design component introduced in Section 2.1; its sufficiency for representing multiomics research is assumed, not independently validated.
  • Conclusion graph (dynamic graph recording validated relationships)
    purpose: Provides long-term memory to guide subsequent iterations and avoids repeating past directions.
    Section 2.1 describes it as the system's long-term record, but no controlled ablation shows the graph alone is responsible for the reported gains.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Automating Exploratory Multiomics Research via Language Models." pith.science (2026). https://pith.science/paper/TOWQDR35

@misc{pith2026250607591,
  author       = {Pith},
  title        = {Pith review of: Automating Exploratory Multiomics Research via Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TOWQDR35}},
  note         = {Machine review of arXiv:2506.07591}
}
read the original abstract

This paper introduces PROTEUS, a fully automated system that produces data-driven hypotheses from raw data files. We apply PROTEUS to clinical proteogenomics, a field where effective downstream data analysis and hypothesis proposal is crucial for producing novel discoveries. PROTEUS uses separate modules to simulate different stages of the scientific process, from open-ended data exploration to specific statistical analysis and hypothesis proposal. It formulates research directions, tools, and results in terms of relationships between biological entities, using unified graph structures to manage complex research processes. We applied PROTEUS to 10 clinical multiomics datasets from published research, arriving at 360 total hypotheses. Results were evaluated through external data validation and automatic open-ended scoring. Through exploratory and iterative research, the system can navigate high-throughput and heterogeneous multiomics data to arrive at hypotheses that balance reliability and novelty. In addition to accelerating multiomic analysis, PROTEUS represents a path towards tailoring general autonomous systems to specialized scientific domains to achieve open-ended hypothesis generation from data.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 42 canonical work pages

  1. [29]

    Publisher: Nature Publishing Group

    Mertins, P., Mani, D.R., Ruggles, K.V ., Gillette, M.A., Clauser, K.R., Wang, P., Wang, X., Qiao, J.W., Cao, S., Petralia, F., Kawaler, E., Mundt, F., Krug, K., Tu, Z., Lei, J.T., Gatza, M.L., Wilkerson, M., Perou, C.M., Yellapantula, V ., Huang, K.-l., Lin, C., McLellan, M.D., Yan, P., Davies, S.R., Townsend, R.R., Skates, S.J., Wang, J., Zhang, B., Kins...

    work page

  2. [33]

    Publisher: Nature Publishing Group

    Stewart, P.A., Welsh, E.A., Slebos, R.J.C., Fang, B., Izumi, V ., Chambers, M., Zhang, G., Cen, L., Pettersson, F., Zhang, Y ., Chen, Z., Cheng, C.-H., Thapa, R., Thompson, Z., Fellows, K.M., Francis, J.M., Saller, J.J., Mesa, T., Zhang, C., Yoder, S., DeNicola, G.M., Beg, A.A., Boyle, T.A., Teer, J.K., Ann Chen, Y ., Koomen, J.M., Eschrich, S.A., Haura, ...

    work page doi:10.1038/s41467-019-11452-x

  3. [1]

    Accessed 2025-02-08

    Rajczewski, A.T., Jagtap, P.D., Griffin, T.J.: An overview of technologies for MS-based proteomics-centric multi-omics 19(3), 165–181 https://doi.org/10.1080/14789450.2022.2070476 . Accessed 2025-02-08

    work page arXiv 2022

  4. [2]

    _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mco2.315

    Chen, C., Wang, J., Pan, D., Wang, X., Xu, Y ., Yan, J., Wang, L., Yang, X., Yang, M., Liu, G.-P.: Applications of multi-omics analysis in human diseases4(4), 315 https://doi.org/10.1002/mco2.315 . _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/mco2.315. Accessed 2025-05-01

    work page doi:10.1002/mco2.315

  5. [3]

    Publisher: Frontiers

    Krassowski, M., Das, V ., Sahu, S.K., Misra, B.B.: State of the field in multi-omics research: From computational needs to data mining and sharing11https://doi.org/10.3389/fgene.2020.610798 . Publisher: Frontiers. Accessed 2025-05-01

    work page arXiv 2020

  6. [4]

    Publisher: Nature Publishing Group

    Cui, M., Cheng, C., Zhang, L.: High-throughput proteomics: a methodological mini-review102(11), 1170–1181 https: //doi.org/10.1038/s41374-022-00830-7 . Publisher: Nature Publishing Group. Accessed 2025-05-01

    work page doi:10.1038/s41374-022-00830-7 2025

  7. [5]

    1002/pmic.202200013

    Messner, C.B., Demichev, V ., Wang, Z., Hartl, J., Kustatscher, G., Mülleder, M., Ralser, M.: Mass spectrometry-based high-throughput proteomics and its role in biomedical studies and systems biology23(7), 2200013 https://doi.org/10. 1002/pmic.202200013 . _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/pmic.202200013. Accessed 2025-05-01

    work page doi:10.1002/pmic.202200013

  8. [6]

    Accessed 2025-05-01

    Yanai, I., Lercher, M.: A hypothesis is a liability21(1), 231 https://doi.org/10.1186/s13059-020-02133-w . Accessed 2025-05-01

    work page doi:10.1186/s13059-020-02133-w 2025

Show all 57 references
  1. [7]

    Accessed 2025-05-01

    Ratti, E.: Big data biology: Between eliminative inferences and exploratory experiments82(2), 198–218 https://doi.org/ 10.1086/680332 . Accessed 2025-05-01

    2025 doi

  2. [8]

    Accessed 2025-05-01

    Yanai, I., Lercher, M.: The data-hypothesis conversation22(1), 58 https://doi.org/10.1186/s13059-021-02277-3 . Accessed 2025-05-01

    2025 doi

  3. [9]

    Accessed 2025-05-01

    Felin, T., Koenderink, J., Krueger, J.I., Noble, D., Ellis, G.F.R.: The data-hypothesis relationship22(1), 57 https://doi.org/ 10.1186/s13059-021-02276-4 . Accessed 2025-05-01

    2025 doi

  4. [10]

    Accessed 2025-05-01

    Mazzocchi, F.: Could big data be the end of theory in science? https://doi.org/10.15252/embr.201541001 . Accessed 2025-05-01

    2025 doi

  5. [11]

    arXiv preprint arXiv:2303.08774 (2023)

    Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S., Anadkat, S., et al.: Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    2023 arXiv

  6. [12]

    arXiv preprint arXiv:2407.21783 (2024)

    Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Letman, A., Mathur, A., Schelten, A., Yang, A., Fan, A., et al.: The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    2024 arXiv

  7. [13]

    AI Open2, 225–250 (2021)

    Han, X., Zhang, Z., Ding, N., Gu, Y ., Liu, X., Huo, Y ., Qiu, J., Yao, Y ., Zhang, A., Zhang, L.,et al.: Pre-trained models: Past, present and future. AI Open2, 225–250 (2021)

    2021

  8. [14]

    arXiv preprint arXiv:2108.07258 (2021)

    Bommasani, R., Hudson, D.A., Adeli, E., Altman, R., Arora, S., Arx, S., Bernstein, M.S., Bohg, J., Bosselut, A., Brunskill, E., et al.: On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258 (2021)

    2021 arXiv

  9. [15]

    arXiv preprint arXiv:2404.18416 (2024)

    Saab, K., Tu, T., Weng, W.-H., Tanno, R., Stutz, D., Wulczyn, E., Zhang, F., Strother, T., Park, C., Vedadi, E., et al.: Capabilities of gemini models in medicine. arXiv preprint arXiv:2404.18416 (2024)

    2024 arXiv

  10. [16]

    arXiv preprint arXiv:2406.03949 (2024)

    Zhang, K., Zeng, S., Hua, E., Ding, N., Chen, Z.-R., Ma, Z., Li, H., Cui, G., Qi, B., Zhu, X., et al.: Ultramedical: Building specialized generalists in biomedicine. arXiv preprint arXiv:2406.03949 (2024)

    2024 arXiv

  11. [17]

    Qin, Y ., Hu, S., Lin, Y ., Chen, W., Ding, N., Cui, G., Zeng, Z., Huang, Y ., Xiao, C., Han, C., Fung, Y .R., Su, Y ., Wang, H., Qian, C., Tian, R., Zhu, K., Liang, S., Shen, X., Xu, B., Zhang, Z., Ye, Y ., Li, B., Tang, Z., Yi, J., Zhu, Y ., Dai, Z., Yan, L., Cong, X., Lu, Y...

    arXiv

  12. [18]

    Bran, A., Cox, S., Schilter, O., Baldassari, C., White, A.D., Schwaller, P.: Augmenting large language models with chemistry tools

    M. Bran, A., Cox, S., Schilter, O., Baldassari, C., White, A.D., Schwaller, P.: Augmenting large language models with chemistry tools. Nature Machine Intelligence

  13. [19]

    arXiv preprint arXiv:2502.09858 (2025)

    Huang, K., Jin, Y ., Li, R., Li, M.Y ., Candès, E., Leskovec, J.: Automated hypothesis validation with agentic sequential falsifications. arXiv preprint arXiv:2502.09858 (2025)

    2025 arXiv

  14. [20]

    bioRxiv 2024.05.13.593861 (2024)

    Xiao, Y ., Liu, J., Zheng, Y ., Xie, X., Hao, J., Li, M., Wang, R., Ni, F., Li, Y ., Luo, J., Jiao, S., Peng, J.: CellAgent: An LLM-driven multi-agent framework for automated single-cell data analysis. bioRxiv 2024.05.13.593861 (2024)

    2024

  15. [21]

    Zhou, J., Zhang, B., Chen, X., Li, H., Xu, X., Chen, S., He, W., Xu, C., Gao, X.: An AI Agent for Fully Automated Automating Exploratory Multiomics Research via Language Models Multi-omic Analyses (2023)

    2023

  16. [22]

    bioRxiv 2024.05.19.594895 (2024)

    Liu, Y ., Shen, R., Zhou, L., Xiao, Q., Yuan, J., Li, Y .: A data-intelligence-intensive bioinformatics copilot system for large-scale omics researches and scientific insights. bioRxiv 2024.05.19.594895 (2024)

    2024

  17. [23]

    bioRxiv 2024.05.22.595240 (2024)

    Xin, Q., Kong, Q., Ji, H.: BioInformatics agent (BIA): Unleashing the power of large language models to reshape bioinformatics workflow. bioRxiv 2024.05.22.595240 (2024)

    2024

  18. [24]

    https://doi.org/10.1101/2024.10.01.616063

    Lu, Y .-C., Varghese, A., Nahar, R., Chen, H., Shao, K., Bao, X., Li, C.: scChat: A Large Language Model-Powered Co-Pilot for Contextualized Single-Cell RNA Sequencing Analysis (2024). https://doi.org/10.1101/2024.10.01.616063

    2024 doi

  19. [25]

    arXiv preprint arXiv:2407.13637 (2024)

    Deng, L., Wu, Y ., Ren, Y ., Lu, H.: DREAM: a biomedical data-driven self-evolving autonomous research system. arXiv preprint arXiv:2407.13637 (2024)

    2024 arXiv

  20. [26]

    Accessed 2024-02-26

    Boiko, D.A., MacKnight, R., Kline, B., Gomes, G.: Autonomous chemical research with large language models624(7992), 570–578 https://doi.org/10.1038/s41586-023-06792-0 . Accessed 2024-02-26

    2024 doi

  21. [27]

    Wang, H., He, Y ., Coelho, P.P., Bucci, M., Nazir, A., Chen, B., Trinh, L., Zhang, S., Huang, K., Chandrasekar, V ., Chung, D.C., Hao, M., Leote, A.C., Lee, Y ., Li, B., Liu, T., Liu, J., Lopez, R., Lucas, T., Ma, M., Makarov, N., McGinnis, L., Peng, L., Ra, S., Scalia, G., Si...

    2025

  22. [28]

    Publisher: American Chemical Society

    Edwards, N.J., Oberti, M., Thangudu, R.R., Cai, S., McGarvey, P.B., Jacob, S., Madhavan, S., Ketchum, K.A.: The CPTAC data portal: A resource for cancer proteomics research14(6), 2707–2713 https://doi.org/10.1021/pr501254j . Publisher: American Chemical Society. Accessed 2025-05-02

    doi

  23. [30]

    Krug, K., Jaehnig, E.J., Satpathy, S., Blumenberg, L., Karpova, A., Anurag, M., Miles, G., Mertins, P., Geffen, Y ., Tang, L.C., Heiman, D.I., Cao, S., Maruvka, Y .E., Lei, J.T., Huang, C., Kothadia, R.B., Colaprico, A., Birger, C., Wang, J., Dou, Y ., Wen, B., Shi, Z., Liao, ...

  24. [31]

    Accessed 2025-03-20

    Zhang, H., Bai, L., Wu, X.-Q., Tian, X., Feng, J., Wu, X., Shi, G.-H., Pei, X., Lyu, J., Yang, G., Liu, Y ., Xu, W., Anwaier, A., Zhu, Y ., Cao, D.-L., Xu, F., Wang, Y ., Gan, H.-L., Sun, M.-H., Zhao, J.-Y ., Qu, Y ., Ye, D., Ding, C.: Pro- teogenomics of clear cell renal cell...

  25. [32]

    Accessed 2025-03-20

    Yanovich-Arad, G., Ofek, P., Yeini, E., Mardamshina, M., Danilevsky, A., Shomron, N., Grossman, R., Satchi-Fainaro, R., Geiger, T.: Proteogenomics of glioblastoma associates molecular patterns with survival34(9), 108787 https://doi.org/ 10.1016/j.celrep.2021.108787 . Accessed ...

  26. [34]

    Accessed 2025-03-20 Automating Exploratory Multiomics Research via Language Models

    Su, H., Chen, L., Wu, J., Cheng, Z., Li, J., Ren, Y ., Xu, J., Dang, Y ., Zheng, M., Cao, Y ., Gao, J., Dai, C., Hu, X., Xie, H., Chen, J., Luo, T., Zhu, J., Wu, C., Sha, W., Chen, C., Liu, H.: Proteogenomic characterization reveals tumorigenesis and progression of lung cancer...

    doi

  27. [35]

    Liu, Q., Zhang, J., Guo, C., Wang, M., Wang, C., Yan, Y ., Sun, L., Wang, D., Zhang, L., Yu, H., Hou, L., Wu, C., Zhu, Y ., Jiang, G., Zhu, H., Zhou, Y ., Fang, S., Zhang, T., Hu, L., Li, J., Liu, Y ., Zhang, H., Zhang, B., Ding, L., Robles, A.I., Rodriguez, H., Gao, D., Ji, H...

    doi

  28. [36]

    Accessed 2025-03-20

    Gao, Q., Zhu, H., Dong, L., Shi, W., Chen, R., Song, Z., Huang, C., Li, J., Dong, X., Zhou, Y ., Liu, Q., Ma, L., Wang, X., Zhou, J., Liu, Y ., Boja, E., Robles, A.I., Ma, W., Wang, P., Li, Y ., Ding, L., Wen, B., Zhang, B., Rodriguez, H., Gao, D., Zhou, H., Fan, J.: Integrate...

    doi

  29. [37]

    Accessed 2025-04-26

    Deng, M., Ran, P., Chen, L., Wang, Y ., Yu, Z., Cai, K., Feng, J., Qin, Z., Yin, Y ., Tan, S., Liu, Y ., Xu, C., Shi, G., Ji, Y ., Zhao, J., Zhou, J., Fan, J., Hou, Y ., Ding, C.: Proteogenomic characterization of cholangiocarcinoma77(2), 411–429 https://doi.org/10.1002/hep.32...

    doi

  30. [38]

    Dong, L., Lu, D., Chen, R., Lin, Y ., Zhu, H., Zhang, Z., Cai, S., Cui, P., Song, G., Rao, D., Yi, X., Wu, Y ., Song, N., Liu, F., Zou, Y ., Zhang, S., Zhang, X., Wang, X., Qiu, S., Zhou, J., Wang, S., Zhang, X., Shi, Y ., Figeys, D., Ding, L., Wang, P., Zhang, B., Rodriguez, ...

    doi

  31. [39]

    Accessed 2025-03-20

    Li, C., Sun, Y .-D., Yu, G.-Y ., Cui, J.-R., Lou, Z., Zhang, H., Huang, Y ., Bai, C.-G., Deng, L.-L., Liu, P., Zheng, K., Wang, Y .-H., Wang, Q.-Q., Li, Q.-R., Wu, Q.-Q., Liu, Q., Shyr, Y ., Li, Y .-X., Chen, L.-N., Wu, J.-R., Zhang, W., Zeng, R.: Integrated omics of metastati...

    doi

  32. [40]

    Monti, S.: Consensus clustering: A resampling-based method for class discovery and visualization of gene expression microarray data

  33. [41]

    Publisher: Elsevier

    Aronow, R.A., Akbarinejad, S., Le, T., Su, S., Shahriyari, L.: TumorDecon: A digital cytometry software18, 101072 https://doi.org/10.1016/j.softx.2022.101072 . Publisher: Elsevier. Accessed 2025-05-02

    2022

  34. [42]

    Publisher: Oxford Academic

    Gong, T., Szustakowski, J.D.: DeconRNASeq: a statistical framework for deconvolution of heterogeneous tissue samples based on mRNA-seq data29(8), 1083–1085 https://doi.org/10.1093/bioinformatics/btt090 . Publisher: Oxford Academic. Accessed 2025-05-02

    doi

  35. [43]

    Publisher: Nature Publishing Group

    Newman, A.M., Liu, C.L., Green, M.R., Gentles, A.J., Feng, W., Xu, Y ., Hoang, C.D., Diehn, M., Alizadeh, A.A.: Robust enumeration of cell subsets from tissue expression profiles12(5), 453–457 https://doi.org/10.1038/nmeth.3337 . Publisher: Nature Publishing Group. Accessed 2025-05-02

    doi

  36. [44]

    Publisher: Oxford Academic

    Fang, Z., Liu, X., Peltz, G.: GSEApy: a comprehensive package for performing gene set enrichment analysis in python 39(1) https://doi.org/10.1093/bioinformatics/btac757 . Publisher: Oxford Academic. Accessed 2025-05-02

    2025 doi

  37. [45]

    Publisher: Oxford Academic

    Ogata, H., Goto, S., Sato, K., Fujibuchi, W., Bono, H., Kanehisa, M.: KEGG: Kyoto encyclopedia of genes and genomes 27(1), 29–34 https://doi.org/10.1093/nar/27.1.29 . Publisher: Oxford Academic. Accessed 2025-05-02

    2025 doi

  38. [46]

    Accessed 2025-05-02

    Kuleshov, M.V ., Xie, Z., London, A.B.K., Yang, J., Evangelista, J.E., Lachmann, A., Shu, I., Torre, D., Ma’ayan, A.: KEA3: improved kinase enrichment analysis via data integration49, 304 https://doi.org/10.1093/nar/gkab359 . Accessed 2025-05-02

    doi

  39. [47]

    https://doi.org/10.5281/zenodo.12549337

    Davidson-Pilon, C.: Lifelines, Survival Analysis in Python. https://doi.org/10.5281/zenodo.12549337 . https://zenodo.org/ records/12549337 Accessed 2024-09-15

    2024

  40. [48]

    Publisher: Oxford Academic

    Szklarczyk, D., Kirsch, R., Koutrouli, M., Nastou, K., Mehryary, F., Hachilif, R., Gable, A.L., Fang, T., Doncheva, N.T., Pyysalo, S., Bork, P., Jensen, L.J., Mering, C.: The STRING database in 2023: protein–protein association networks and functional enrichment analyses for a...

    2023 doi

  41. [49]

    S/T/YXXX_ProteinName

    Badia-i-Mompel, P., Vélez Santiago, J., Braunger, J., Geiss, C., Dimitrov, D., Müller-Dott, S., Taus, P., Dugourd, A., Holland, C.H., Ramirez Flores, R.O., Saez-Rodriguez, J.: decoupleR: ensemble of computational methods to infer bio- logical activities from omics data2(1) htt...

    doi

  42. [50]

    **Tissue type**: Whether the sample is from tumor tissue or tumor-adjacent normal tissue (tumor, normal)

  43. [51]

    **Metastases**: Metastases situation of the tumor (M0, M1)

  44. [52]

    **Status at diagnosis**: Tumor status at the time of diagnosis (Localized, Advanced)

  45. [53]

    **Stage at diagnosis**: Tumor stage at the time of diagnosis as classified by the TNM system (I, II, III, IV)

  46. [54]

    **RECIST**: Tumor RECIST classification (SD, PR, PD, CR)

  47. [55]

    **Smoking**: Whether the patient smokes (Y es, No)

  48. [56]

    **Live Status**: Survival event indicator (1 = dead, 0 = alive)

  49. [57]

    Table B1: Full hypotheses proposed by the system, corresponding to the two case studies

    **OS**: Overall survival time in months B Case Study Hypotheses Table B1 lists the original hypotheses produced byPROTEUSthat were analyzed as case studies. Table B1: Full hypotheses proposed by the system, corresponding to the two case studies. Dataset Name Hypothesis ccRCC H...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.