MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models

Ahmed Abdeen Hamed; Bhanwar Lal Puniya; Josh Loecker; Narayna Puraja; Tom\'a\v{s} Helikar; William Bryan

REVIEW 5 major objections 4 minor 66 references

MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models

T0 review · 5 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read MechAInistic claims that an LLM-based multi-agent system can turn plain-English biology questions into executable, traceable workflows over genome-scale metabolic models, yielding auditable drug-target hypotheses.

desk verdict A genuinely useful, honestly reported multi-agent system for constraint-based metabolic modeling, but the evidence for the central 'converts' claim rests on two single runs with manual correction and no public code. read the letter →

arxiv 2607.18249 v1 pith:3KU7FBUO submitted 2026-05-08 q-bio.QM cs.AI

Josh Loecker , Narayna Puraja , William Bryan , Bhanwar Lal Puniya , Ahmed Abdeen Hamed , Tomáš Helikar This is my paper

classification q-bio.QMcs.AI

keywords agenticAIworkfloworchestrationmetabolicmodelingconstraint-basedmechanisticdrugrepurposingfluxbalanceanalysistherapeutichypothesisgeneration

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

MechAInistic claims that large language models become trustworthy for metabolic-model reasoning when their outputs are forced through a fixed registry of executable modeling tools and a reviewer agent that checks each step. The paper builds a three-agent system—an Architect that plans tool calls, a Reviewer that scores plans and evidence, and a Task agent that summarizes outputs and searches PubMed—and shows it converting a plain-English drug-repurposing question into a concrete workflow on paired disease/healthy metabolic models. In two immune-cell use cases the system nominated Devimistat/CPI-613 against OGDH in rheumatoid arthritis naive B cells and ivosidenib against IDH in multiple sclerosis Th17 cells, each choice backed by flux-distance metrics and simulated perturbations. The authors deliberately treat the drug identities as hypotheses, not validated answers; the headline result is that a natural-language question can be turned into an auditable chain of model-derived numbers, perturbation simulations, and literature support.

What carries the argument

The load-bearing mechanism is the separation of planning from verification. Three role-specialized agents—Architect, Reviewer, and Task—operate on a session state that stores the uploaded models, the user query, and a complete record of tool calls. The Architect proposes tool calls drawn from a fixed registry of COBRApy-backed functions (flux-balance simulation, flux-distance calculation, reaction/pathway comparison, gene/reaction knockout, dose-response, drug-target lookup, PubMed retrieval); the Reviewer scores plans and intermediate evidence on a numeric scale, forcing regeneration when scores are low; the Task agent summarizes raw JSON tool outputs into plain text, cutting token usage by

What would settle it

Construct two steady-state models that are identical except for a single known reaction knockout or a single known drug-target perturbation, run MechAInistic with the standard prompt, and check whether the system's highest-ranked single target is that known reaction; an exhaustive in-silico screen can then verify whether the recommended target actually minimizes flux distance to the target state.

Watch

Extended reading notes

Core claim

The central claim is that workflow structure, not language-model size, is what makes LLM-based reasoning trustworthy: MechAInistic's Architect-Reviewer-Task loop keeps biological conclusions grounded in executable constraint-based modeling rather than free-form text. Concretely, the paper shows the system taking the same prompt—identify a single drug therapy that restores disease-state flux to the healthy state while minimizing off-target disruption—and, in both the RA and MS use cases, loading the JSON models, computing baseline flux distances, prioritizing reactions that carry disease-only flux, simulating inhibition, and retrieving literature before naming a therapy. The system also outpe

Load-bearing premise

The disease-state metabolic models, reconstructed from very few bulk RNA-seq samples (4 for RA, 19 for MS) via COMO on Recon3D, faithfully capture the metabolic state of those immune cells, so that flux differences from the healthy reference reflect disease biology rather than reconstruction artifacts.

Editorial extensions

If this is right

A biologist who can phrase a question in English can run a full constraint-based-model drug-target screen without writing code, provided they can supply paired current/target models.
The same Architect-Reviewer-Task workflow transfers to any pair of COBRA-compatible models, so the method is not tied to the two immune-cell use cases shown.
Comparisons of LLM-based analysis tools should be scored on traceability and execution audit, not on the plausibility of the final narrative or the identity of the recommended drug.
Because the system keeps quantitative outputs in the report, hypotheses can be independently re-run and audited—a prerequisite for using AI-generated biology in a lab setting.
The paper's own caveats imply that the limiting factor for hypothesis quality becomes the input metabolic model, not the agent software.

Reading between the lines

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

If the Architect-Reviewer pattern generalizes, the same grounding trick could be applied to other simulation-backed biological questions, such as kinetic models or agent-based immune models, not just constraint-based metabolism.
The flux-distance-minimization objective used here could serve as a community benchmark: a documented library of paired models with known interventions would let anyone measure whether agentic LLM tools recover the known target.
Because LLM outputs vary across runs and providers, practical adoption will likely require pinned model versions, deterministic seeds, and local caching of external database calls; the paper reports configurations but does not promise bit-for-bit reproducibility.
The RA and MS reconstructions rest on 4 and 19 bulk RNA-seq samples respectively; if those models are not faithful, the mechanism-level claims become artifacts of the reconstruction, which suggests that agent tools may push validation effort upstream into model building.

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.

Desk Editor's Note

read the letter

You should know this paper before the field floods with LLM-agent metabolic tools: MechAInistic is one of the more serious attempts, and it is not just a wrapper around ChatGPT. The Architect-Reviewer-Task architecture is sensible, the tool registry is real (COBRApy, HiGHS, flux distance, perturbation, DGIdb, PubMed), and the authors did the right thing by pitting general-purpose LLMs against it on the same models and prompt. Their audit of Claude showing plausible narrative hiding execution-level errors is a nice piece of evidence for why tool grounding matters.

What is genuinely new: applying a reviewer-supervised agentic loop specifically to paired constraint-based models and producing two concrete, model-grounded drug hypotheses (Devimistat/CPI-613 for RA naive B cells, ivosidenib for MS Th17 cells). The paper is also refreshingly honest about its own failure modes: a citation mismatch, an annotation error (ICDHy), the need for manual correction of the final drug, and the explicit statement that the disease models require separate validation. That honesty is real and should be credited.

The soft spots are the ones the stress test names. Each use case is a single run at temperature 0.5 with no seed, no complete tool-call transcript, and no public code repository. The authors manually validated and corrected the outputs after generation, so the "traceable" workflow is partly a post hoc curation. The RA model comes from four RNA-seq samples from one study; the MS model from nineteen. The nine-axis rubric was scored by the same team that built the system. None of this is fatal on its own, but together it means the headline claim—"converts natural-language questions into executable, model-grounded workflows"—is supported more as a demonstration of concept than as a reproducible system property. The authors themselves acknowledge LLM variability across runs, which undercuts any single-run demonstration.

My bottom line: this deserves serious peer review, not desk rejection. The architecture and the comparator methodology are worth the field's attention. But the referees should push hard for public code, multiple independent runs, and raw tool logs; without those, the central claim remains anecdotal. As it stands, I would not yet cite it as a validated system, but I would tell people to watch it.

Send it to review, with a request for substantial revision and reproducibility evidence.

Referee Report

5 major / 4 minor

Summary. The paper presents MechAInistic, a multi-agent LLM-guided system for reasoning over paired genome-scale constraint-based metabolic models. The system uses an Architect-Reviewer-Task architecture to translate a natural-language biological question into an executable, tool-grounded workflow, then produces a structured report with flux-based quantitative evidence, literature support, and explicit limitations. The authors demonstrate the system on two therapeutic hypothesis-generation use cases: RA naive B cells (recommending Devimistat/CPI-613 targeting OGDH) and MS CD4+ Th17 cells (recommending ivosidenib targeting IDH1/IDH2). They also compare MechAInistic with general-purpose LLM chatbots using a nine-axis rubric, reporting that MechAInistic achieves better model grounding and traceability. The manuscript is transparent about several limitations, including dependence on input model quality, variability of LLM behavior across runs, the need for manual validation, and documented citation-level errors.

Significance. If the central claim is supported, MechAInistic would be a valuable contribution: it lowers the barrier to constraint-based metabolic modeling, provides an auditable chain from user query to model-derived results, and demonstrates a meaningful use of multi-agent LLM systems for scientific reasoning. The architecture is clearly described, the tool registry is sensible, and the authors explicitly frame their recommendations as model-grounded hypotheses rather than validated therapies. The honesty about failure modes and the inclusion of two real biological use cases are strengths. However, the current evidence is closer to a pilot demonstration than a validated system claim: each use case is run once, the outputs are manually curated, and the central claim that the system 'converts' a natural-language question into a workflow is therefore not yet established with the reproducibility expected for a methods paper.

major comments (5)

[Results 'Model-grounded workflows'; Methods 'LLM configuration and prompt management'] The two use-case demonstrations are each a single execution at temperature 0.5 with no random seed or model-version pinning reported in the main text. The paper's own Discussion states that LLM behavior can vary across runs, so a single success does not establish that MechAInistic reliably 'converts' natural-language questions into workflows. Public code, full tool-call transcripts, and independent reruns are needed; without these, the central claim in the Abstract is anecdotal. I ask the authors to report at least 3-5 runs per use case, with the variance in workflow structure and recommendations, and to release logs and code.
[Methods 'Manual validation of MechAInistic outputs'; Results use cases] The final reports appear to have been produced after manual validation that corrected a title-author mismatch, an ICDHy annotation error, and the final drug choice (vorasidenib over ivosidenib). The manuscript does not show the raw pre-validation system outputs or a diff of what human reviewers changed. Because the central claim is that the system itself generates traceable reports, the reader must be able to distinguish system output from curated post-processing. Please include raw outputs or an explicit change log.
[Methods 'Comparator LLM evaluation' and 'Nine-axis evaluation rubric'] The primary evidence for improved grounding over general-purpose LLMs is a nine-axis rubric scored qualitatively by the authors, with no blinding and no inter-rater reliability assessment. Since the same group designed MechAInistic and the rubric, the comparison is vulnerable to confirmation bias. Please provide independent scoring, or at minimum raw comparator outputs and detailed per-axis justifications, to support the claim of 'most consistent workflow-level performance'.
[Methods 'Metabolic model reconstruction and use-case setup'] Reference 41 is cited as the source of the RA bulk RNA-seq data, but the reference is Wang et al. (Nat. Commun. 2018) on IL-21 in SLE, not RA. Reference 38 is cited for the healthy Naive B model but is the COMO pipeline preprint. These provenance errors prevent readers from reproducing the input models and weaken the biological use cases. Please verify all accessions/references and, if the text is correct, add the GEO accessions so the models can be reconstructed.
[Results use cases; Methods 'Constraint-based modeling backend'] The system operationalizes 'therapeutic efficacy' as reduction of Euclidean flux distance to the healthy state (e.g., 25.247 to 23.069 cited as '8.6% improvement'). This is an unvalidated modeling assumption, not a model-derived theorem; different distance metrics or objective functions could change the ranking of targets and drugs. The paper should explicitly label this assumption and include sensitivity analysis across distance measures (Euclidean, Manhattan, cosine, etc.) to support the robustness of the recommended hypotheses.

minor comments (4)

[Figure 2D / Results Evaluation phase] The phrase 'the all summarized tool call results' is ungrammatical; should be 'all the summarized tool call results'.
[Throughout] Capitalization of COBRApy is inconsistent ('CobraPy' appears in the Constraint-based modeling backend section); please use the standard 'COBRApy'.
[Table 3 and Results] Model versions such as 'Kimi K2.6, Qwen3.5, Gemma-4' should include access dates and provider endpoints; it is also not clear in the main text which agent was assigned to which model.
[Abstract vs. Results] The Abstract states that MechAInistic 'proposed ivosidenib as an FDA-approved repurposing candidate,' but the Results note that human review suggested vorasidenib may be mechanistically preferable. Please qualify the Abstract to make clear that ivosidenib is the system's output and that expert review altered the compound-level recommendation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported outputs are declared model-grounded hypotheses and follow from the supplied models and explicit analysis tools, not from back-fitted parameters or load-bearing self-citations.

full rationale

MechAInistic's central claim is that it converts a natural-language query into an executable workflow over two supplied COBRA JSON models. The quantitative outputs (Euclidean flux distances, cosine similarity, reaction fluxes, biomass objective, perturbation effects) are computed by named tools (COBRApy, HiGHS, FBA, production-envelope robustness, knockout/dose-response simulations) and are not fit parameters later renamed as predictions. The therapeutic nominations (Devimistat/CPI-613, ivosidenib) are openly framed as model-grounded hypotheses, not validated predictions: the paper states "These outputs should not be interpreted as validated therapeutic recommendations; rather, they are model-grounded hypotheses for expert review and experimental validation." The selection of OGDH and ICDHy is defined by the same explicit criteria (disease-state flux with no healthy flux, simulated distance reduction), so the reported improvement is the algorithm's objective function, not an independent prediction; the paper presents this transparently as a model-derived simulation result. The same-lab citations (COMO, AutoRNAseq, prior healthy-model papers) are pipeline/tool references, and the paper itself limits their role: "the models themselves require separate documentation and validation." No uniqueness theorem, ansatz-via-citation, or renaming is used. The limitations that do exist — a single un-replicated run per use case at temperature 0.5, manual validation after generation, identified citation/annotation errors, and no public code or full tool-call transcripts — are reproducibility and external-validity concerns, not circularity of the derivation chain.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No numeric parameters are fitted to make the central claim succeed; the system instead relies on standard CBM assumptions, the fidelity of small-sample reconstructions, and the choice of Euclidean flux distance as a health-restoration metric. The three agents are software roles, not new scientific entities.

assumptions (3)

domain assumption Flux balance analysis steady-state solutions represent biologically meaningful metabolic states.
Standard in constraint-based modeling; the entire toolset relies on FBA as the primary simulation paradigm.
domain assumption The disease-state models reconstructed from limited RNA-seq data (RA: 4 samples; MS: 19 samples) are accurate enough to identify disease-relevant metabolic differences.
All flux comparisons and drug-target hypotheses depend on this fidelity; the paper itself notes the models 'require separate documentation and validation.'
ad hoc to paper Euclidean distance between flux distributions measures the degree of metabolic divergence, and reducing this distance toward the healthy state corresponds to therapeutic benefit.
This metric is a modeling choice in the tool; no biological validation is provided for treating Euclidean flux distance as the objective of therapy.

how reviews work

0 comments

Cite this review

Pith. "Pith review of MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models." pith.science (2026). https://pith.science/paper/3KU7FBUO

@misc{pith2026260718249,
  author       = {Pith},
  title        = {Pith review of: MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3KU7FBUO}},
  note         = {Machine review of arXiv:2607.18249}
}

read the original abstract

Constraint-based metabolic modeling is a powerful way to study the mechanistic basis of cellular states and disease, but its effective use demands substantial computational expertise and careful coordination of multi-step analyses. We developed MechAInistic to lower this barrier and enable researchers to ask complex biological questions in natural language. Harnessing large language models, MechAInistic is a multi-agent system organized around an Architect-Reviewer pattern that transforms a natural-language question into an executable, model-grounded workflow and generates a structured report. The system supports a variety of tasks, including pathway comparison, perturbation analysis, drug-target exploration, and literature-grounded interpretation across paired metabolic model states. We developed and evaluated MechAInistic using two paired immune-cell metabolic-model use cases for therapeutic hypothesis generation. For Naive B cells from rheumatoid arthritis (RA) paired with healthy controls, MechAInistic identified mitochondrial metabolic rewiring and nominated Devimistat/CPI-613 as an investigational OGDH-centered hypothesis. In a paired CD4+ Th17 cell study from multiple sclerosis (MS) and healthy controls, the same workflow identified NADP-dependent isocitrate dehydrogenase as the optimal single target and proposed ivosidenib as an FDA-approved repurposing candidate. Together, these results show that MechAInistic converts natural-language biological questions into executable, model-grounded workflows for traceable therapeutic hypothesis generation.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

66 extracted references · 4 canonical work pages

[2]

M., King, Z

Bordbar, A., Monk, J. M., King, Z. A. & Palsson, B. O. Constraint-based models predict metabolic and associated cellular functions. Nat. Rev. Genet. 15 , 107–120 (2014)

2014
[3]

Lubbock, A. L. R. & Lopez, C. F. Programmatic modeling for biological systems. bioRxiv (2024)

2024
[4]

Computational evidence: inclusion of flux values, perturbation outputs, distance metrics, dose-response results, or other quantitative model-derived outputs
[5]

Direct relevance: adherence to the single-drug flux-restoration task rather than drifting into general disease biology or unrelated therapeutic framing
[6]

Question completion: whether the system returned a specific single therapy in response to the prompt
[7]

Structured review or agentic control: evidence of a review, checkpointing, or multi-agent control process before final synthesis
[8]

Execution feasibility and rerunnability: completion without file-ingestion failure, truncation, unavailable context, or undocumented execution constraints
[9]

Each axis was scored qualitatively as full, partial, or failed based on pre-specified criteria

Human-review traceability: explicit indication of manual review, sign-off, correction, or identifiable points where human review could be applied. Each axis was scored qualitatively as full, partial, or failed based on pre-specified criteria. Drug identity was recorded descriptively but was not used as the main correctness criterion, because the computati...

2025

Show all 66 references

[10]

Puniya, B. L. & others. Perspectives on computational modeling of biological systems and the significance of the SysMod community. Bioinforma. Adv. 4 , vbae090 (2024)

2024
[11]

& others

Heirendt, L. & others. Creation and analysis of biochemical constraint-based models using the COBRA Toolbox v3.0. Nat. Protoc. 14 , 639–702 (2019)

2019
[12]

& others

Lapi, F. & others. COBRAxy: constraint-based metabolic modeling in Galaxy. Bioinformatics (2025)

2025
[13]

A., Palsson, B

Ebrahim, A., Lerman, J. A., Palsson, B. O. & Hyduke, D. R. COBRApy: COnstraints-Based Reconstruction and Analysis for Python. BMC Syst. Biol. 7 , 74 (2013)

2013
[14]

& Jolicoeur, M

Yasemi, M. & Jolicoeur, M. Modelling Cell Metabolism: A Review on Constraint-Based Steady-State and Kinetic Approaches. Processes 9 , 467 (2021)

2021
[15]

Benchmarking Constraint-Based Metabolic Models: How to Validate FBA Predictions Against Experimental Flux Data

Bennett, O. Benchmarking Constraint-Based Metabolic Models: How to Validate FBA Predictions Against Experimental Flux Data. PlantChemica (2026)

2026
[16]

& others

Ng, C. & others. Constraint-Based Reconstruction and Analyses of Metabolic Models: Open-Source Python Tools and Applications to Cancer. Front. Oncol. 12 , 829524 (2022)

2022
[17]

& Herrgård, M

Machado, D. & Herrgård, M. J. Systematic evaluation of methods for integration of transcriptomic data into constraint-based models. PLoS Comput. Biol. 10 , e1003580 (2014)

2014
[18]

& others

Brunk, E. & others. Recon3D enables a three-dimensional view of gene variation in human metabolism. Cell Syst. 6 , 1–14 (2018)

2018
[19]

W., & others

Wehling, L., Singh, G., Mulyadi, A. W., & others. Talk2Biomodels: AI agent-based open-source LLM initiative for kinetic biological models. BMC Bioinformatics 26 , 1–14 (2025)

2025
[20]

& Roy, S

Alam, T. & Roy, S. From Prompt to Pipeline: Large Language Models for Scientific Workflow Development in Bioinformatics. bioRxiv (2025)

2025
[21]

& others

Rao, R. & others. Generalist biological artificial intelligence in modeling the language of life. Nat. Biotechnol. (2026)

2026
[22]

Philip, P. A. et al. A Phase III Open-Label Trial to Evaluate Efficacy and Safety of CPI-613 Plus Modified FOLFIRINOX (MFFX) versus FOLFIRINOX (FFX) in Patients with Metastatic Adenocarcinoma of the Pancreas. Future Oncol. 15 , 3189–3196 (2019)

2019
[23]

L., Culos, K

Merchant, S. L., Culos, K. & Wyatt, H. Ivosidenib: IDH1 Inhibitor for the Treatment of Acute Myeloid Leukemia. J. Adv. Pract. Oncol. 10 , 494–500 (2019)

2019
[24]

Metformin-Induced Mitochondrial Complex I Inhibition: Facts, Uncertainties, and Consequences

Fontaine, E. Metformin-Induced Mitochondrial Complex I Inhibition: Facts, Uncertainties, and Consequences. Front. Endocrinol. 9 , 753 (2018)

2018
[25]

Reinhart, G. A. Vitamin D analogs: novel therapeutic agents for cardiovascular disease? Curr. Opin. Investig. Drugs 5 , 947–951 (2004)

2004
[26]

Nakken, B. et al. B-cells and their targeting in rheumatoid arthritis — Current concepts and future perspectives. Autoimmun. Rev. 11 , 28–34 (2011)

2011
[27]

Nguyen, T. T. et al. OGDH and Bcl-xL loss causes synthetic lethality in glioblastoma. JCI Insight 9 , e172565 (2024)

2024
[28]

Udumula, M. P. et al. Targeting mitochondrial metabolism with CPI-613 in chemoresistant ovarian tumors. J. Ovarian Res. 17 , 226 (2024)

2024
[29]

Khan, H. Y. et al. Targeting Cellular Metabolism With CPI-613 Sensitizes Pancreatic Cancer Cells to Radiation Therapy. Adv. Radiat. Oncol. 8 , 101122 (2023)

2023
[30]

Reddy, V. B. et al. In Vitro and In Vivo Metabolism of a Novel Antimitochondrial Cancer Metabolism Agent, CPI-613, in Rat and Human. Drug Metab. Dispos. Biol. Fate Chem. 50 , 361–373 (2022)

2022
[31]

Xu, R. et al. Potassium ion efflux induces exaggerated mitochondrial damage and non-pyroptotic necrosis when energy metabolism is blocked. Free Radic. Biol. Med. 212 , 117–132 (2024)

2024
[32]

& Ziemssen, T

Moser, T., Akgün, K., Proschmann, U., Sellner, J. & Ziemssen, T. The role of TH17 cells in multiple sclerosis: Therapeutic implications. Autoimmun. Rev. 19 , 102647 (2020)

2020
[33]

Puniya, B. L. et al. A Mechanistic Computational Model Reveals That Plasticity of CD4+ T Cell Differentiation Is a Function of Cytokine Composition and Dosage. Front. Physiol. 9 , 878 (2018)

2018
[34]

Fang, Y. et al. Isocitrate Dehydrogenase Mutations in Cancer: From Bench to Bedside Applications. https://doi.org/10.1002/mco2.70732 doi:10.1002/mco2.70732

doi
[35]

Ding, Y.-H. et al. Inhibition of SLC25A10 promotes cellular senescence and impedes hepatocellular carcinoma progression. Transl. Cancer Res. 14 , 4939–4954 (2025)

2025
[36]

& Zhong, S

Anselme, M., He, H., Lai, C., Luo, W. & Zhong, S. Targeting mitochondrial transporters and metabolic reprogramming for disease treatment. J. Transl. Med. 23 , 1111 (2025)

2025
[37]

Xu, T. et al. Metabolic control of TH17 and induced Treg cell balance by an epigenetic mechanism. Nature 548 , 228–233 (2017)

2017
[38]

Mellinghoff, I. K. et al. Vorasidenib in IDH1- or IDH2-Mutant Low-Grade Glioma. N. Engl. J. Med. 389 , 589–601 (2023)

2023
[39]

Wagner, A. et al. Metabolic Modeling of Single Th17 Cells Reveals Regulators of Autoimmunity. Cell 184 , 4168-4185.e21 (2021)

2021
[40]

Kaushik, D. K. & Yong, V. W. Metabolic needs of brain-infiltrating leukocytes and microglia in multiple sclerosis. J. Neurochem. 158 , 14–24 (2021)

2021
[41]

Weitzel, D. et al. The National Research Platform: Stretched, Multi-Tenant, Scientific Kubernetes Cluster. in Practice and Experience in Advanced Research Computing 2025: The Power of Collaboration 1–5 (Association for Computing Machinery, New York, NY, USA, 2025). doi:10.1145...

2025
[42]

& Hall, J

Huangfu, Q. & Hall, J. A. J. Parallelizing the dual revised simplex method. Math. Program. Comput. 10 , 119–142 (2018)

2018
[43]

S., Ramakrishna, R

Edwards, J. S., Ramakrishna, R. & Palsson, B. O. Characterizing the metabolic phenotype: A phenotype phase plane analysis. Biotechnol. Bioeng. 77 , 27–36 (2002)

2002
[44]

Bessell, B. et al. COMO: A Pipeline for Multi-Omics Data Integration in Metabolic Modeling and Drug Discovery. Preprint at https://doi.org/10.48550/arXiv.2011.02103 (2023)

2011 doi arXiv
[45]

Puniya, B. L. et al. Integrative computational approach identifies drug targets in CD4+ T-cell-mediated immune disorders. Npj Syst. Biol. Appl. 7 , 4 (2021)

2021
[46]

Clough, E. et al. NCBI GEO: archive for gene expression and epigenomics data sets: 23-year update. Nucleic Acids Res. 52 , D138–D144 (2023)

2023
[47]

Wang, S. et al. IL-21 drives expansion and plasma cell differentiation of autoreactive CD11chiT-bet+ B cells in SLE. Nat. Commun. 9 , 1758 (2018)

2018
[48]

Loecker, J., Bessell, B., Puniya, B. L. & Helikar, T. AutoRNAseq: Automated Bulk RNA-seq Analysis Pipeline. 2026.04.21.719844 Preprint at https://doi.org/10.64898/2026.04.21.719844 (2026)

2026 doi
[49]

Dyer, S. C. et al. Ensembl 2025. Nucleic Acids Res. 53 , D948–D957 (2025)

2025
[50]

Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29 , 15–21 (2013)

2013
[51]

I., Irizarry, R

Patro, R., Duggal, G., Love, M. I., Irizarry, R. A. & Kingsford, C. Salmon provides fast and bias-aware quantification of transcript expression. Nat. Methods 14 , 417–419 (2017)

2017
[52]

Norsigian, C. J. et al. BiGG Models 2020: multi-strain genome-scale models and expansion across the phylogenetic tree. Nucleic Acids Res. 48 , D402–D406 (2020)

2020
[53]

Cannon, M. et al. DGIdb 5.0: rebuilding the drug–gene interaction database for precision medicine and drug discovery platforms. Nucleic Acids Res. 52 , D1227–D1235 (2024)

2024
[54]

Xin, J. et al. High-performance web services for querying gene and variant annotation. Genome Biol. 17 , 91 (2016)

2016
[55]

Sayers, E. W. et al. Database resources of the National Center for Biotechnology Information in 2025. Nucleic Acids Res. 53 , D20–D29 (2025)

2025
[56]

K., Hernandez, J

Li, B., Saini, A. K., Hernandez, J. G. & Moore, J. H. Agentic AI and the rise of in silico team science in biomedical research. Nat. Biotechnol. 1–15 (2026) doi:10.1038/s41587-026-03035-1

2026 doi
[57]

Zhao, W. et al. An agentic system for rare disease diagnosis with traceable reasoning. Nature 651 , 775–784 (2026)

2026
[58]

Gao, Z. et al. Examining the Potential of ChatGPT on Biomedical Information Retrieval: Fact-Checking Drug-Disease Associations. Ann. Biomed. Eng. 52 , 1919–1927 (2024)

1919
[59]

Wang, Z. et al. GeneAgent: self-verification language agent for gene-set analysis using domain databases. Nat. Methods 22 , 1677–1685 (2025)

2025
[60]

Hamed, A. A. & Rocha, L. M. Protocol for Evaluating ChatGPT in Biomedical Association Generation and Verification Using a RAG ‑ Enabled, Cross ‑ Model Majority Voting Workflow. SSRN Scholarly Paper at https://doi.org/10.2139/ssrn.5366821 (2025)

2025 doi
[61]

A., Crimi, A., Misiak, M

Hamed, A. A., Crimi, A., Misiak, M. M. & Lee, B. S. From knowledge generation to knowledge verification: examining the biomedical generative capabilities of ChatGPT. iScience 28 , (2025)

2025
[62]

A., Zachara-Szymanska, M

Hamed, A. A., Zachara-Szymanska, M. & Wu, X. Safeguarding authenticity for mitigating the harms of generative AI: Issues, research agenda, and policies for detection, fact-checking, and ethical AI. iScience 27 , (2024)

2024
[63]

& Buettner, R

Woesle, C., Fischer-Brandies, L. & Buettner, R. A Systematic Literature Review of Hallucinations in Large Language Models. IEEE Access 13 , 148231–148253 (2025)

2025
[64]

Ji, Z. et al. Survey of Hallucination in Natural Language Generation. ACM Comput Surv 55 , 248:1-248:38 (2023)

2023
[65]

& McDonald, R

Maynez, J., Narayan, S., Bohnet, B. & McDonald, R. On Faithfulness and Factuality in Abstractive Summarization. in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (eds Jurafsky, D., Chai, J., Schluter, N. & Tetreault, J.) 1906–1919 (Asso...

1906 doi
[66]

& Roberts, K

Hsu, E. & Roberts, K. Leveraging large language models for knowledge-free weak supervision in clinical natural language processing. Sci. Rep. 15 , 8241 (2025)

2025
[67]

Bzdok, D. et al. Data science opportunities of large language models for neuroscience and biomedicine. Neuron 112 , 698–717 (2024)

2024

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [2]

M., King, Z

Bordbar, A., Monk, J. M., King, Z. A. & Palsson, B. O. Constraint-based models predict metabolic and associated cellular functions. Nat. Rev. Genet. 15 , 107–120 (2014)

2014

[2] [3]

Lubbock, A. L. R. & Lopez, C. F. Programmatic modeling for biological systems. bioRxiv (2024)

2024

[3] [4]

Computational evidence: inclusion of flux values, perturbation outputs, distance metrics, dose-response results, or other quantitative model-derived outputs

[4] [5]

Direct relevance: adherence to the single-drug flux-restoration task rather than drifting into general disease biology or unrelated therapeutic framing

[5] [6]

Question completion: whether the system returned a specific single therapy in response to the prompt

[6] [7]

Structured review or agentic control: evidence of a review, checkpointing, or multi-agent control process before final synthesis

[7] [8]

Execution feasibility and rerunnability: completion without file-ingestion failure, truncation, unavailable context, or undocumented execution constraints

[8] [9]

Each axis was scored qualitatively as full, partial, or failed based on pre-specified criteria

Human-review traceability: explicit indication of manual review, sign-off, correction, or identifiable points where human review could be applied. Each axis was scored qualitatively as full, partial, or failed based on pre-specified criteria. Drug identity was recorded descriptively but was not used as the main correctness criterion, because the computati...

2025