Pith. sign in

REVIEW 4 major objections 4 minor 99 references

An Auditable Agent Platform For Automated Molecular Optimisation

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a hierarchical multi-agent LLM system with auditable provenance improves focused binding optimization, raising average predicted binding affinity against AKT1 by 31%.

desk verdict The auditability mechanism is the real contribution; the 31% headline claim is unverifiable from the supplied text and likely conflates architectural benefit with a larger docking-tool budget. read the letter →

arxiv 2508.03444 v1 pith:PVU5WLZO submitted 2025-08-05 cs.LG

Atabey Ünlü , Phil Rohr , Ahmet Celebi This is my paper
classification cs.LG
keywords molecularoptimizationmulti-agentLLMdockingscoreprovenancerecordsAKT1drugdiscoverytest-timescalingauditableAI
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 attempts to show that a hierarchical multi-agent LLM system with tool calls and recorded provenance can automate molecular optimization against a target protein. In runs against AKT1, the multi-agent configuration improved average predicted binding affinity by 31%, outperforming single-agent and unguided LLM configurations on that metric. The paper also identifies a trade-off: single-agent runs produced molecules with better drug-like properties while binding less strongly, and unguided runs were fastest but left their reasoning paths unverified. The central claim is that test-time scaling, focused feedback loops, and provenance records turn general-purpose LLMs into auditable molecular design systems.

What carries the argument

The carrying mechanism is the provenance record: concise records that capture molecular lineage and summarize each tool call, so agents can reuse successful transformations through in-context learning and the full reasoning path stays inspectable. Around this record sits a hierarchy of six LLM agents with defined roles, including a Medicinal Chemist that invokes a docking tool during edits and a Ranking agent that scores candidates by predicted binding affinity. The docking score is both steering signal and evaluation metric, making it the object that carries the optimization loop.

What would settle it

Run the top molecules from each configuration through an experimental binding assay on AKT1, such as surface plasmon resonance or a cellular IC50 test; if the 31% multi-agent advantage in docking scores does not reproduce in measured affinities, or the ordering across configurations changes, the central claim is refuted.

Watch

Extended reading notes

Core claim

The paper's central discovery is that orchestrating multiple specialized LLM agents—a Principal Researcher, Database agent, AI Expert, Medicinal Chemist, Ranking agent, and Scientific Critic—with explicit tool invocations and provenance-carrying records improves focused binding optimization over both a single agent and an unguided model. Steering edits through a docking tool, scoring candidates, and policing logic let the multi-agent team raise average predicted binding affinity against AKT1 by 31% in the authors' scale-up runs. The single-agent configuration traded some binding potency for superior drug-like properties, while unguided LLM runs were fastest but did not expose the tool signals needed to verify reasoning. The paper interprets this as evidence that test-time scaling and feedback loops, rather than model architecture improvements, are what convert general-purpose LLMs into useful and auditable design tools.

Load-bearing premise

The argument depends on the docking tool's predicted binding affinity being a meaningful proxy for real binding, since the same score both guides the editing edits and measures success.

Editorial extensions

If this is right

  • If the central claim holds, multi-agent orchestration with docking feedback is a practical way to improve binding scores without retraining or fine-tuning the underlying LLMs.
  • Auditable provenance means each proposed molecule can be traced back through its editing history, which supports failure analysis and reporting in drug discovery settings.
  • The trade-off between binding potency and drug-like properties suggests that a production configuration might combine multi-agent binding optimization with single-agent property filtering.
  • Extending the same tool-use pattern to ADMET or selectivity predictors, as the paper suggests, could push automated workflows beyond hit-finding toward lead optimization.

Reading between the lines

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

  • Editorial extension: a direct test of whether the 31% gain comes from multi-agent dialogue or simply from more LLM calls would hold the docking tool fixed and vary only the number of agents against a fixed compute budget.
  • Editorial extension: if the docking score were replaced by an experimental assay as the feedback signal, the same agent architecture should transfer without changing agent roles, with assay throughput becoming the bottleneck.
  • Editorial extension: the stored provenance trajectories could be used as training examples to distill multi-agent behavior into a single smaller model, preserving much of the audit trail at lower inference cost.
  • Editorial extension: the reported advantage is on one target, so a natural next check is whether the same configuration reproduces the 31%-scale improvement on a panel of unrelated proteins before treating it as a general property of multi-agent design.
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

4 major / 4 minor

Summary. The paper describes a hierarchical multi-agent LLM platform for automated molecular optimization, with roles for a Principal Researcher, Database agent, AI Expert, Medicinal Chemist, Ranking agent, and Scientific Critic, and claims that provenance records make the reasoning paths inspectable. Three research loops were run against the AKT1 protein using five LLMs, followed by 20 independent scale-ups on the two top-scoring models, and comparisons across LLM-only, single-agent, and multi-agent configurations. The central claim is that the multi-agent configuration improves average predicted binding affinity by 31% over the other configurations, while single-agent runs produce better drug-like properties and unguided runs are fastest but less auditable. The full text supplied to me is mostly illegible, so the verification of these claims rests largely on the abstract and fragments.

Significance. If the claims hold, the paper would provide a useful demonstration that test-time scaling, focused docking feedback, and explicit provenance can convert general-purpose LLMs into auditable molecular-design agents. The explicit provenance and inspectability design is a valuable direction, and the comparison across three agent configurations addresses a question of current interest. However, the significance is currently bounded by a single protein target, a single in silico scoring oracle, and the absence of any external validation, so the contribution is more incremental than transformative until the computational claims are made verifiable.

major comments (4)
  1. [Abstract, central claim] The 31% improvement in average predicted binding affinity is reported as a point estimate with no variance, sample size, significance test, or explicit definition of the baseline. Because the protocol first ranks five LLMs by mean docking score and then runs 20 scale-ups on only the two top performers, the improvement may be partly selection-driven if the comparison includes the other three models or if the 'leading LLMs' are chosen post hoc. Please report per-model means with standard deviations or confidence intervals, the number of independent runs per configuration, and a paired or mixed-effects test that accounts for model identity.
  2. [Abstract, experimental protocol and compute fairness] The abstract states that the Medicinal Chemist agent invokes a docking tool while editing and that the Ranking agent scores candidates, so the multi-agent arm is structurally expected to perform more docking-tool calls than the LLM-only or single-agent arms. If the number of docking invocations, LLM calls, or wall-clock time is not matched across arms, the 31% gain could reflect a larger optimization budget applied to the same objective rather than an architectural benefit. Please report per-configuration counts of docking calls, LLM calls, and wall-clock time, and include a control that gives the LLM-only configuration the same number of oracle calls or an equivalent compute budget.
  3. [Abstract, outcome metric and single-scorer optimization] The outcome measure is predicted binding affinity from a docking tool, and the same type of tool is used by the Medicinal Chemist to guide edits and by the Ranking agent to score candidates. Optimizing and evaluating with the same function can inflate apparent gains even when the objective is a pre-existing external function, so the 31% improvement may partly reflect overfitting to that one scorer. Please validate the generated molecules with a different docking or scoring function (or experimental data, if available), and report distributional properties such as scaffold diversity, synthetic accessibility, and drug-likeness to show that the multi-agent gain is not obtained by exploiting a single scoring function's weaknesses.
  4. [Full text, tables and statistics] The full text supplied to me is heavily corrupted and illegible; the tables, figure axes, and result paragraphs cannot be read, so none of the per-arm numbers, variances, docking-tool call counts, or statistical tests could be verified. This is not merely a typographical issue because the central claim depends on these numbers. A revised submission must provide a readable manuscript with complete tables (or a machine-readable statistics and data appendix) that reports the exact experimental protocol and all quantitative results before the claim can be fairly evaluated.
minor comments (4)
  1. [Abstract] The terms 'three cycle research loops' and '20 independent scale ups' are not defined in the abstract; please specify what one 'cycle' and one 'scale up' consists of, including the number of generated molecules and the stopping criterion.
  2. [Methods, models and tools] The 'sequence to molecule deep learning model' and the docking tool are not identified by name or version; naming them (for example, the specific generative model and docking program with its scoring function version) is important for reproducibility.
  3. [Auditability claims] The abstract claims that the full reasoning path remains inspectable, but no concrete example of a provenance record is shown; including a sample provenance trace or a figure illustrating the molecule lineage and tool-call sequence would make this claim testable.
  4. [Terminology] The phrase 'architectural trade off' appears to conflate the agent architecture with the orchestration of tool calls and compute budgets; please define precisely what distinguishes the three configurations (which agents are active, which tools are available, and how many calls each configuration may make).

Circularity Check

1 steps flagged · score 4.0 of 10

Multi-agent gain is reported on the same docking score that steers the loop, making the 31% improvement a partially closed-loop result.

  1. other [Abstract: agent-roles sentence and results paragraph]
    "a Medicinal Chemist edits them while invoking a docking tool, a Ranking agent scores the candidates ... After ranking the models by mean docking score, we ran 20 independent scale ups on the two top performers ... the multi agent setting excelled at focused binding optimization, improving average predicted binding affinity by 31%."

    The reported outcome ('average predicted binding affinity') is produced by the same docking tool that the Medicinal Chemist invokes during editing and that the Ranking agent uses to score candidates. The evaluation metric is therefore identical, by construction, to the feedback signal steering the multi-agent loop. The 31% gain is an in-sample measure of how well the agents optimized the docking scorer, not an independent prediction of binding affinity. Because the docking tool is an external, pre-existing function rather than a parameter fitted by these authors, the result is not definitionally forced; but the central comparison partially reduces to optimizing and evaluating with one function.

full rationale

The paper's central claim is an experimental outcome rather than a formal derivation, and no self-citation chains or imported uniqueness theorems are visible in the readable portions. The main circularity-adjacent issue is that the docking score acts as both the optimizer's steering signal and the reported evaluation metric: the Medicinal Chemist invokes the docking tool while editing, the Ranking agent scores candidates with it, and the headline 31% improvement is stated as 'average predicted binding affinity.' This is a closed-loop evaluation design, so the reported gain is partly an artifact of optimizing and measuring with the same function. However, the docking scorer is a fixed external oracle, not a parameter fitted to the test molecules, so the result is not equivalent to the input by definition; the architectural comparison still contains independent empirical content. The corrupted full text prevented verification of per-arm docking-call counts, error bars, or additional methodological details, so the score of 4 reflects the closed-loop evaluation rather than a stronger definitional circularity.

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

No fitted constants or invented entities appear in the abstract; the central claim rests on protocol and tooling assumptions. The three axioms listed are domain assumptions about the docking score's validity, the stability of the five-model ranking, and the comparability of the three configurations. None of these can be confirmed from the abstract, and the garbled full text prevents checking them.

assumptions (3)
  • domain assumption The docking tool's predicted binding affinity is a valid proxy for true binding affinity, so optimizing it constitutes meaningful binding optimization.
    The abstract reports 'predicted binding affinity' as the outcome and the docking tool as the steering signal; if the scorer does not track real binding, the 31% gain is a scoring artifact.
  • domain assumption The ranking of the five LLMs by mean docking score is stable enough that scaling up the top two does not inflate the reported improvement.
    The abstract describes ranking five models and running 20 scale-ups on the top two; with noisy per-model scores, selecting winners by the same metric then re-measuring on it biases the gain upward (regression to the mean).
  • domain assumption The three configurations (LLM-only, single-agent, multi-agent) differ only in architecture, with comparable compute and evaluation protocol.
    The abstract compares the configurations but gives no detail on token budgets, run counts per configuration, or evaluation standardization, so the trade-off could reflect resource differences.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Auditable Agent Platform For Automated Molecular Optimisation." pith.science (2026). https://pith.science/paper/PVU5WLZO

@misc{pith2026250803444,
  author       = {Pith},
  title        = {Pith review of: An Auditable Agent Platform For Automated Molecular Optimisation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PVU5WLZO}},
  note         = {Machine review of arXiv:2508.03444}
}
read the original abstract

Drug discovery frequently loses momentum when data, expertise, and tools are scattered, slowing design cycles. To shorten this loop we built a hierarchical, tool using agent framework that automates molecular optimisation. A Principal Researcher defines each objective, a Database agent retrieves target information, an AI Expert generates de novo scaffolds with a sequence to molecule deep learning model, a Medicinal Chemist edits them while invoking a docking tool, a Ranking agent scores the candidates, and a Scientific Critic polices the logic. Each tool call is summarised and stored causing the full reasoning path to remain inspectable. The agents communicate through concise provenance records that capture molecular lineage, to build auditable, molecule centered reasoning trajectories and reuse successful transformations via in context learning. Three cycle research loops were run against AKT1 protein using five large language models. After ranking the models by mean docking score, we ran 20 independent scale ups on the two top performers. We then compared the leading LLMs' binding affinity results across three configurations, LLM only, single agent, and multi agent. Our results reveal an architectural trade off, the multi agent setting excelled at focused binding optimization, improving average predicted binding affinity by 31%. In contrast, single agent runs generated molecules with superior drug like properties at the cost of less potent binding scores. Unguided LLM runs finished fastest, yet their lack of transparent tool signals left the validity of their reasoning paths unverified. These results show that test time scaling, focused feedback loops and provenance convert general purpose LLMs into auditable systems for molecular design, and suggest that extending the toolset to ADMET and selectivity predictors could push research workflows further along the discovery pipeline.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

99 extracted references · 50 canonical work pages

  1. [1]

    Machine learning-aided generative molecular design

    Yuanqi Du, Arian R Jamasb, Jeff Guo, Tianfan Fu, Charles Harris, Yingheng Wang, Chenru Duan, Pietro Li \`o , Philippe Schwaller, and Tom L Blundell. Machine learning-aided generative molecular design. Nature Machine Intelligence , 6(6):589--604, 2024

    2024

  2. [2]

    From understanding diseases to drug design: can artificial intelligence bridge the gap? Artificial Intelligence Review , 57(4):86, 2024

    Anju Choorakottayil Pushkaran and Alya A Arabi. From understanding diseases to drug design: can artificial intelligence bridge the gap? Artificial Intelligence Review , 57(4):86, 2024

    2024

  3. [3]

    Application of artificial intelligence in drug design: A review

    Simrandeep Singh, Navjot Kaur, and Anita Gehlot. Application of artificial intelligence in drug design: A review. Computers in Biology and Medicine , 179:108810, 2024

    2024

  4. [4]

    A survey of generative ai for de novo drug design: new frontiers in molecule and protein generation

    Xiangru Tang, Howard Dai, Elizabeth Knight, Fang Wu, Yunyang Li, Tianxiao Li, and Mark Gerstein. A survey of generative ai for de novo drug design: new frontiers in molecule and protein generation. Briefings in Bioinformatics , 25(4), 2024

    2024

  5. [5]

    Machine-learning-assisted molecular design of innovative polymers

    Tianle Yue, Jianxin He, and Ying Li. Machine-learning-assisted molecular design of innovative polymers. Accounts of Materials Research , 2025

    2025

  6. [6]

    Orchestrating tool ecosystem of drug discovery with intention-aware llm agents

    Mingyu Derek Ma, Karina Zadorozhny, Jesse Swanson, Nathan C Frey, Keunwoo Choi, Maksim Eremeev, Sabrina J Mielke, Wenmo Sun, Melody Liu, Jonathan Wickes, et al. Orchestrating tool ecosystem of drug discovery with intention-aware llm agents. In Towards Agentic AI for Science: Hypothesis Generation, Comprehension, Quantification, and Validation

  7. [7]

    Drugpilot: Llm-based parameterized reasoning agent for drug discovery

    Kun Li, Zhennan Wu, Shoupeng Wang, and Wenbin Hu. Drugpilot: Llm-based parameterized reasoning agent for drug discovery. arXiv preprint arXiv:2505.13940 , 2025

    arXiv 2025

  8. [8]

    Ai-enabled language models (lms) to large language models (llms) and multimodal large language models (mllms) in drug discovery and development

    Chiranjib Chakraborty, Manojit Bhattacharya, Soumen Pal, Srijan Chatterjee, Arpita Das, and Sang-Soo Lee. Ai-enabled language models (lms) to large language models (llms) and multimodal large language models (mllms) in drug discovery and development. Journal of Advanced Research , 2025

    2025

Show all 99 references
  1. [9]

    Artificial intelligence in drug development

    Kang Zhang, Xin Yang, Yifei Wang, Yunfang Yu, Niu Huang, Gen Li, Xiaokun Li, Joseph C Wu, and Shengyong Yang. Artificial intelligence in drug development. Nature medicine , 31(1):45--59, 2025

    2025

  2. [10]

    A breakthrough in r&d for neglected diseases: new ways to get the drugs we need

    Mary Moran. A breakthrough in r&d for neglected diseases: new ways to get the drugs we need. PLoS medicine , 2(9):e302, 2005

    2005

  3. [11]

    Ai with agency: a vision for adaptive, efficient, and ethical healthcare

    Vasco Gerardo Hinostroza Fuentes, Hezerul Abdul Karim, Myles Joshua Toledo Tan, and Nouar AlDahoul. Ai with agency: a vision for adaptive, efficient, and ethical healthcare. Frontiers in Digital Health , 7:1600216, 2025

    2025

  4. [12]

    Drug target selection and validation

    Marcus T Scotti and Carolina L Bellera. Drug target selection and validation . Springer, 2022

    2022

  5. [13]

    Pharmacophore modeling and its applications

    Rashmi Tyagi, Amisha Singh, Kamal Kumar Chaudhary, and Manoj Kumar Yadav. Pharmacophore modeling and its applications. In Bioinformatics , pages 269--289. Elsevier, 2022

    2022

  6. [14]

    Peptide-based drug discovery through artificial intelligence: towards an autonomous design of therapeutic peptides

    Montserrat Goles, Anamar \' a Daza, Gabriel Cabas-Mora, Lindybeth Sarmiento-Var \'o n, Julieta Sep \'u lveda-Ya \ n ez, Hoda Anvari-Kazemabad, Mehdi D Davari, Roberto Uribe-Paredes, \'A lvaro Olivera-Nappa, Marcelo A Navarrete, et al. Peptide-based drug discovery through artif...

    2024

  7. [15]

    Artificial intelligence and machine learning-aided drug discovery in central nervous system diseases: State-of-the-arts and future directions

    Sezen Vatansever, Avner Schlessinger, Daniel Wacker, H \"U mit Kaniskan, Jian Jin, Ming-Ming Zhou, and Bin Zhang. Artificial intelligence and machine learning-aided drug discovery in central nervous system diseases: State-of-the-arts and future directions. Medicinal research r...

    2021

  8. [16]

    Ai-driven drug discovery for rare diseases

    Amit Gangwal and Antonio Lavecchia. Ai-driven drug discovery for rare diseases. Journal of Chemical Information and Modeling , 65(5):2214--2231, 2024

    2024

  9. [17]

    Patient-centric knowledge graphs: a survey of current methods, challenges, and applications

    Hassan S Al Khatib, Subash Neupane, Harish Kumar Manchukonda, Noorbakhsh Amiri Golilarz, Sudip Mittal, Amin Amirlatifi, and Shahram Rahimi. Patient-centric knowledge graphs: a survey of current methods, challenges, and applications. Frontiers in Artificial Intelligence , 7:138...

    2024

  10. [18]

    In-context learning with long-context models: An in-depth exploration

    Amanda Bertsch, Maor Ivgi, Emily Xiao, Uri Alon, Jonathan Berant, Matthew R Gormley, and Graham Neubig. In-context learning with long-context models: An in-depth exploration. arXiv preprint arXiv:2405.00200 , 2024

    2024 arXiv

  11. [19]

    An information-theoretic analysis of in-context learning

    Hong Jun Jeon, Jason D Lee, Qi Lei, and Benjamin Van Roy. An information-theoretic analysis of in-context learning. arXiv preprint arXiv:2401.15530 , 2024

    2024 arXiv

  12. [20]

    Synergpt: In-context learning for personalized drug synergy prediction and drug design

    Carl Edwards, Aakanksha Naik, Tushar Khot, Martin Burke, Heng Ji, and Tom Hope. Synergpt: In-context learning for personalized drug synergy prediction and drug design. arXiv preprint arXiv:2307.11694 , 2023

    2023 arXiv

  13. [21]

    A survey on in-context learning

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Tianyu Liu, et al. A survey on in-context learning. arXiv preprint arXiv:2301.00234 , 2022

    2022 arXiv

  14. [22]

    Improving drug-drug interaction prediction via in-context learning and judging with large language models

    He Qi, Xiaoqiang Li, Chengcheng Zhang, and Tianyi Zhao. Improving drug-drug interaction prediction via in-context learning and judging with large language models. Frontiers in Pharmacology , 16:1589788, 2025

    2025

  15. [23]

    Mhnfs: Prompting in-context bioactivity predictions for low-data drug discovery

    Johannes Schimunek, Sohvi Lukkonen, and Gunter Klambauer. Mhnfs: Prompting in-context bioactivity predictions for low-data drug discovery. Journal of Chemical Information and Modeling , 65(9):4243--4250, 2025

    2025

  16. [24]

    The virtual lab: Ai agents design new sars-cov-2 nanobodies with experimental validation

    Kyle Swanson, Wesley Wu, Nash L Bulaong, John E Pak, and James Zou. The virtual lab: Ai agents design new sars-cov-2 nanobodies with experimental validation. bioRxiv , pages 2024--11, 2024

    2024

  17. [25]

    The ai scientist: Towards fully automated open-ended scientific discovery

    Chris Lu, Cong Lu, Robert Tjarko Lange, Jakob Foerster, Jeff Clune, and David Ha. The ai scientist: Towards fully automated open-ended scientific discovery. arXiv preprint arXiv:2408.06292 , 2024

    2024 arXiv

  18. [26]

    Towards an ai co-scientist

    Juraj Gottweis, Wei-Hung Weng, Alexander Daryin, Tao Tu, Anil Palepu, Petar Sirkovic, Artiom Myaskovsky, Felix Weissenberger, Keran Rong, Ryutaro Tanno, et al. Towards an ai co-scientist. arXiv preprint arXiv:2502.18864 , 2025

    2025 arXiv

  19. [27]

    Alphaevolve: A coding agent for scientific and algorithmic discovery

    Alexander Novikov, Ng \^a n V \ u , Marvin Eisenberger, Emilien Dupont, Po-Sen Huang, Adam Zsolt Wagner, Sergey Shirobokov, Borislav Kozlovskii, Francisco JR Ruiz, Abbas Mehrabian, et al. Alphaevolve: A coding agent for scientific and algorithmic discovery. arXiv preprint arXi...

    2025 arXiv

  20. [28]

    A survey on test-time scaling in large language models: What, how, where, and how well? arXiv preprint arXiv:2503.24235 , 2025

    Qiyuan Zhang, Fuyuan Lyu, Zexu Sun, Lei Wang, Weixu Zhang, Wenyue Hua, Haolun Wu, Zhihan Guo, Yufei Wang, Niklas Muennighoff, et al. A survey on test-time scaling in large language models: What, how, where, and how well? arXiv preprint arXiv:2503.24235 , 2025

    2025 arXiv

  21. [29]

    Large language models in drug discovery: A comprehensive analysis of drug-target interaction prediction

    Raghad J AbuNasser, Mostafa Z Ali, Yaser Jararweh, Mustafa Daraghmeh, and Talal Z Ali. Large language models in drug discovery: A comprehensive analysis of drug-target interaction prediction. In 2024 2nd International Conference on Foundation and Large Language Models (FLLM) ,...

    2024

  22. [30]

    Augmenting dmta using predictive ai modelling at astrazeneca

    Gian Marco Ghiandoni, Emma Evertsson, David J Riley, Christian Tyrchan, and Prakash Chandra Rathi. Augmenting dmta using predictive ai modelling at astrazeneca. Drug discovery today , 29(4):103945, 2024

    2024

  23. [31]

    Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D White, and Philippe Schwaller

    Andres M. Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D White, and Philippe Schwaller. Augmenting large language models with chemistry tools. Nature Machine Intelligence , 6(5):525--535, 2024

    2024

  24. [32]

    Aviary: training language agents on challenging scientific tasks

    Siddharth Narayanan, James D Braza, Ryan-Rhys Griffiths, Manu Ponnapati, Albert Bou, Jon Laurent, Ori Kabeli, Geemi Wellawatte, Sam Cox, Samuel G Rodriques, et al. Aviary: training language agents on challenging scientific tasks. arXiv preprint arXiv:2412.21154 , 2024

    2024 arXiv

  25. [33]

    Toolformer: Language models can teach themselves to use tools

    Timo Schick, Jane Dwivedi-Yu, Roberto Dess \` , Roberta Raileanu, Maria Lomeli, Eric Hambro, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer: Language models can teach themselves to use tools. Advances in Neural Information Processing Systems , 36:68539--68551, 2023

    2023

  26. [34]

    Api-blend: A comprehensive corpora for training and benchmarking api llms

    Kinjal Basu, Ibrahim Abdelaziz, Subhajit Chaudhury, Soham Dan, Maxwell Crouse, Asim Munawar, Sadhana Kumaravel, Vinod Muthusamy, Pavan Kapanipathi, and Luis A Lastras. Api-blend: A comprehensive corpora for training and benchmarking api llms. arXiv preprint arXiv:2402.15491 , 2024

    2024 arXiv

  27. [35]

    Toolllm: Facilitating large language models to master 16000+ real-world apis

    Yujia Qin, Shihao Liang, Yining Ye, Kunlun Zhu, Lan Yan, Yaxi Lu, Yankai Lin, Xin Cong, Xiangru Tang, Bill Qian, et al. Toolllm: Facilitating large language models to master 16000+ real-world apis. arXiv preprint arXiv:2307.16789 , 2023

    2023 arXiv

  28. [36]

    Mathsensei: a tool-augmented large language model for mathematical reasoning

    Debrup Das, Debopriyo Banerjee, Somak Aditya, and Ashish Kulkarni. Mathsensei: a tool-augmented large language model for mathematical reasoning. arXiv preprint arXiv:2402.17231 , 2024

    2024 arXiv

  29. [37]

    Art: Automatic multi-step reasoning and tool-use for large language models

    Bhargavi Paranjape, Scott Lundberg, Sameer Singh, Hannaneh Hajishirzi, Luke Zettlemoyer, and Marco Tulio Ribeiro. Art: Automatic multi-step reasoning and tool-use for large language models. arXiv preprint arXiv:2303.09014 , 2023

    2023 arXiv

  30. [38]

    Cactus: Chemistry agent connecting tool usage to science

    Andrew D McNaughton, Gautham Krishna Sankar Ramalaxmi, Agustin Kruel, Carter R Knutson, Rohith A Varikoti, and Neeraj Kumar. Cactus: Chemistry agent connecting tool usage to science. ACS omega , 9(46):46563--46573, 2024

    2024

  31. [39]

    Sciagent: Tool-augmented language models for scientific reasoning

    Yubo Ma, Zhibin Gou, Junheng Hao, Ruochen Xu, Shuohang Wang, Liangming Pan, Yujiu Yang, Yixin Cao, Aixin Sun, Hany Awadalla, et al. Sciagent: Tool-augmented language models for scientific reasoning. arXiv preprint arXiv:2402.11451 , 2024

    2024 arXiv

  32. [40]

    Pharmagents: Building a virtual pharma with large language model agents

    Bowen Gao, Yanwen Huang, Yiqiao Liu, Wenxuan Xie, Wei-Ying Ma, Ya-Qin Zhang, and Yanyan Lan. Pharmagents: Building a virtual pharma with large language model agents. arXiv preprint arXiv:2503.22164 , 2025

    2025 arXiv

  33. [41]

    Protagents: protein discovery via large language model multi-agent collaborations combining physics and machine learning

    Alireza Ghafarollahi and Markus J Buehler. Protagents: protein discovery via large language model multi-agent collaborations combining physics and machine learning. Digital Discovery , 3(7):1389--1409, 2024

    2024

  34. [42]

    Agent laboratory: Using llm agents as research assistants

    Samuel Schmidgall, Yusheng Su, Ze Wang, Ximeng Sun, Jialian Wu, Xiaodong Yu, Jiang Liu, Zicheng Liu, and Emad Barsoum. Agent laboratory: Using llm agents as research assistants. arXiv preprint arXiv:2501.04227 , 2025

    2025 arXiv

  35. [43]

    A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges

    Xinyi Li, Sai Wang, Siqi Zeng, Yu Wu, and Yi Yang. A survey on llm-based multi-agent systems: workflow, infrastructure, and challenges. Vicinagearth , 1(1):9, 2024

    2024

  36. [44]

    Llm multi-agent systems: Challenges and open problems

    Shanshan Han, Qifan Zhang, Yuhang Yao, Weizhao Jin, and Zhaozhuo Xu. Llm multi-agent systems: Challenges and open problems. arXiv preprint arXiv:2402.03578 , 2024

    2024 arXiv

  37. [45]

    A survey on context-aware multi-agent systems: techniques, challenges and future directions

    Hung Du, Srikanth Thudumu, Rajesh Vasa, and Kon Mouzakis. A survey on context-aware multi-agent systems: techniques, challenges and future directions. arXiv preprint arXiv:2402.01968 , 2024

    2024 arXiv

  38. [46]

    Cooperative intelligent control of multi-agent systems (mas) through communication, trust, and reliability

    Stefalo Acha and Sun Yi. Cooperative intelligent control of multi-agent systems (mas) through communication, trust, and reliability. Control Systems and Optimization Letters , 3(1):53--67, 2025

    2025

  39. [47]

    Multi-agent systems and their applications

    Jing Xie and Chen-Ching Liu. Multi-agent systems and their applications. Journal of International Council on Electrical Engineering , 7(1):188--197, 2017

    2017

  40. [48]

    Online planning for multi-agent systems with bounded communication

    Feng Wu, Shlomo Zilberstein, and Xiaoping Chen. Online planning for multi-agent systems with bounded communication. Artificial Intelligence , 175(2):487--511, 2011

    2011

  41. [49]

    Multi-agent coordination across diverse applications: A survey

    Lijun Sun, Yijun Yang, Qiqi Duan, Yuhui Shi, Chao Lyu, Yu-Cheng Chang, Chin-Teng Lin, and Yang Shen. Multi-agent coordination across diverse applications: A survey. arXiv preprint arXiv:2502.14743 , 2025

    2025 arXiv

  42. [50]

    Advancing multi-agent systems through model context protocol: Architecture, implementation, and applications

    Naveen Krishnan. Advancing multi-agent systems through model context protocol: Architecture, implementation, and applications. arXiv preprint arXiv:2504.21030 , 2025

    2025 arXiv

  43. [51]

    Agent-oriented planning in multi-agent systems

    Ao Li, Yuexiang Xie, Songze Li, Fugee Tsung, Bolin Ding, and Yaliang Li. Agent-oriented planning in multi-agent systems. arXiv preprint arXiv:2410.02189 , 2024

    2024 arXiv

  44. [52]

    Multi-agent collaboration mechanisms: A survey of llms

    Khanh-Tung Tran, Dung Dao, Minh-Duong Nguyen, Quoc-Viet Pham, Barry O'Sullivan, and Hoang D Nguyen. Multi-agent collaboration mechanisms: A survey of llms. arXiv preprint arXiv:2501.06322 , 2025

    2025 arXiv

  45. [53]

    Autogen: Enabling next-gen llm applications via multi-agent conversations

    Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Beibin Li, Erkang Zhu, Li Jiang, Xiaoyun Zhang, Shaokun Zhang, Jiale Liu, et al. Autogen: Enabling next-gen llm applications via multi-agent conversations. In First Conference on Language Modeling , 2024

    2024

  46. [54]

    Tom Taulli and Gaurav Deshmukh. Crewai. In Building Generative AI Agents: Using LangGraph, AutoGen, and CrewAI , pages 103--145. Springer, 2025

    2025

  47. [55]

    Multi-agent systems: A survey about its components, framework and workflow

    Diego Maldonado, Edison Cruz, Jackeline Abad Torres, Patricio J Cruz, and Silvana del Pilar Gamboa Benitez. Multi-agent systems: A survey about its components, framework and workflow. IEEE Access , 12:80950--80975, 2024

    2024

  48. [56]

    A hierarchical design framework for distributed control of multi-agent systems

    Xiangyu Wang, Yujing Xu, Yue Cao, and Shihua Li. A hierarchical design framework for distributed control of multi-agent systems. Automatica , 160:111402, 2024

    2024

  49. [57]

    Agentnet: Decentralized evolutionary coordination for llm-based multi-agent systems

    Yingxuan Yang, Huacan Chai, Shuai Shao, Yuanyi Song, Siyuan Qi, Renting Rui, and Weinan Zhang. Agentnet: Decentralized evolutionary coordination for llm-based multi-agent systems. arXiv preprint arXiv:2504.00587 , 2025

    2025 arXiv

  50. [58]

    Pc-agent: A hierarchical multi-agent collaboration framework for complex task automation on pc

    Haowei Liu, Xi Zhang, Haiyang Xu, Yuyang Wanyan, Junyang Wang, Ming Yan, Ji Zhang, Chunfeng Yuan, Changsheng Xu, Weiming Hu, et al. Pc-agent: A hierarchical multi-agent collaboration framework for complex task automation on pc. arXiv preprint arXiv:2502.14282 , 2025

    2025 arXiv

  51. [59]

    A dynamic llm-powered agent network for task-oriented agent collaboration

    Zijun Liu, Yanzhe Zhang, Peng Li, Yang Liu, and Diyi Yang. A dynamic llm-powered agent network for task-oriented agent collaboration. In First Conference on Language Modeling , 2024

    2024

  52. [60]

    Scaling large language model-based multi-agent collaboration

    Chen Qian, Zihao Xie, Yifei Wang, Wei Liu, Kunlun Zhu, Hanchen Xia, Yufan Dang, Zhuoyun Du, Weize Chen, Cheng Yang, et al. Scaling large language model-based multi-agent collaboration. arXiv preprint arXiv:2406.07155 , 2024

    2024 arXiv

  53. [61]

    Nucleic acids research , 53(D1):D609--D617, 2025

    Uniprot: the universal protein knowledgebase in 2025. Nucleic acids research , 53(D1):D609--D617, 2025

    2025

  54. [62]

    Updated resources for exploring experimentally-determined pdb structures and computed structure models at the rcsb protein data bank

    Stephen K Burley, Rusham Bhatt, Charmi Bhikadiya, Chunxiao Bi, Alison Biester, Pratyoy Biswas, Sebastian Bittrich, Santiago Blaumann, Ronald Brown, Henry Chao, et al. Updated resources for exploring experimentally-determined pdb structures and computed structure models at the ...

    2025

  55. [63]

    Fifteen years of chembl and its role in cheminformatics and drug discovery

    Barbara Zdrazil. Fifteen years of chembl and its role in cheminformatics and drug discovery. Journal of Cheminformatics , 17(1):1--9, 2025

    2025

  56. [64]

    Rdkit documentation

    Greg Landrum. Rdkit documentation. Release , 1(1–79):4, 2013

    2013

  57. [65]

    A. Ünlü, E. Çevrim, A.F. Öztürk, and T. Doğan. Prot2Mol : Target based molecule generation using protein embeddings and SELFIES molecule representation. https://github.com/HUBioDataLab/Prot2Mol, 2024. GitHub repository

    2024

  58. [66]

    Autodock vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading

    Oleg Trott and Arthur J Olson. Autodock vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. Journal of computational chemistry , 31(2):455--461, 2010

    2010

  59. [67]

    Plip: fully automated protein--ligand interaction profiler

    Sebastian Salentin, Sven Schreiber, V Joachim Haupt, Melissa F Adasme, and Michael Schroeder. Plip: fully automated protein--ligand interaction profiler. Nucleic acids research , 43(W1):W443--W447, 2015

    2015

  60. [68]

    Alphafold protein structure database in 2024: providing structure coverage for over 214 million protein sequences

    Mihaly Varadi, Damian Bertoni, Paulyna Magana, Urmila Paramval, Ivanna Pidruchna, Malarvizhi Radhakrishnan, Maxim Tsenkov, Sreenath Nair, Milot Mirdita, Jingi Yeo, et al. Alphafold protein structure database in 2024: providing structure coverage for over 214 million protein se...

    2024

  61. [69]

    Quantifying the chemical beauty of drugs

    G Richard Bickerton, Gaia V Paolini, J \'e r \'e my Besnard, Sorel Muresan, and Andrew L Hopkins. Quantifying the chemical beauty of drugs. Nature chemistry , 4(2):90--98, 2012

    2012

  62. [70]

    Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions

    Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. Journal of cheminformatics , 1:1--11, 2009

    2009

  63. [71]

    Can large language models explore in-context? Advances in Neural Information Processing Systems , 37:120124--120158, 2024

    Akshay Krishnamurthy, Keegan Harris, Dylan J Foster, Cyril Zhang, and Aleksandrs Slivkins. Can large language models explore in-context? Advances in Neural Information Processing Systems , 37:120124--120158, 2024

    2024

  64. [72]

    Transfer-prompting: Enhancing cross-task adaptation in large language models via dual-stage prompts optimization

    Yupeng Chang, Yi Chang, and Yuan Wu. Transfer-prompting: Enhancing cross-task adaptation in large language models via dual-stage prompts optimization. arXiv preprint arXiv:2502.14211 , 2025

    2025 arXiv

  65. [73]

    Evaluating the prompt steerability of large language models

    Erik Miehling, Michael Desmond, Karthikeyan Natesan Ramamurthy, Elizabeth M Daly, Pierre Dognin, Jesus Rios, Djallel Bouneffouf, and Miao Liu. Evaluating the prompt steerability of large language models. arXiv preprint arXiv:2411.12405 , 2024

    2024 arXiv

  66. [74]

    Comparative analysis of prompt strategies for large language models: Single-task vs

    Manuel Gozzi and Federico Di Maio. Comparative analysis of prompt strategies for large language models: Single-task vs. multitask prompts. Electronics , 13(23):4712, 2024

    2024

  67. [75]

    Marche, Hervé Lerat, and Zuzana Macek Jílková

    Mariam Mroweh, Gaël Roth, Thomas Decaens, Patrice N. Marche, Hervé Lerat, and Zuzana Macek Jílková. Targeting AKT in hepatocellular carcinoma and its tumor microenvironment. International Journal of Molecular Sciences , 22(4):1794, 2021

    2021

  68. [76]

    Manning and Lewis C

    Brendan D. Manning and Lewis C. Cantley. AKT / PKB signaling: Navigating downstream. Cell , 129(7):1261--1274, June 2007

    2007

  69. [77]

    Matheny Jr and Martin L

    Ronald W. Matheny Jr and Martin L. Adamo. Current perspectives on AKT `` AKT -ivation'' and `` AKT -ions''. Experimental Biology and Medicine , 234(11):1264--1270, 2009

    2009

  70. [78]

    AKT inhibitors: New weapons in the fight against breast cancer? Frontiers in Pharmacology , 12:662232, 2021

    Federica Martorana, Gianmarco Motta, Giuliana Pavone, Lucia Motta, Stefania Stella, Silvia Rita Vitale, Livia Manzella, and Paolo Vigneri. AKT inhibitors: New weapons in the fight against breast cancer? Frontiers in Pharmacology , 12:662232, 2021

    2021

  71. [79]

    AKT1 transcriptomic landscape in breast cancer cells

    Bijesh George, Bin Gui, Rajeswari Raguraman, Aswathy Mary Paul, Harikrishna Nakshatri, Madhavan Radhakrishna Pillai, and Rakesh Kumar. AKT1 transcriptomic landscape in breast cancer cells. Cells , 11(15):2290, 2022

    2022

  72. [80]

    Integrating molecular design resources within modern drug discovery research: The Roche experience

    Martin Stahl, Wolfgang Guba, and Manfred Kansy. Integrating molecular design resources within modern drug discovery research: The Roche experience. Drug Discovery Today , 11(7--8):326--333, 2006

    2006

  73. [81]

    Differential regulation of AKT1 contributes to survival and proliferation in hepatocellular carcinoma cells by mediating NOTCH1 expression

    Jing Chen, Jun Liang, Shihai Liu, Shanai Song, Wenxuan Guo, and Fangzhen Shen. Differential regulation of AKT1 contributes to survival and proliferation in hepatocellular carcinoma cells by mediating NOTCH1 expression. Oncology Letters , 15(5):6857--6864, 2018

    2018

  74. [82]

    Thomas, Maria Deak, Dario R

    Christine C. Thomas, Maria Deak, Dario R. Alessi, and Daan M. F. van Aalten. High-resolution structure of the pleckstrin homology domain of protein kinase b/ AKT bound to phosphatidylinositol (3,4,5)-trisphosphate. Current Biology , 12(14):1256--1262, 2002

    2002

  75. [83]

    Turner, Mafalda Oliveira, Sacha J

    Nicholas C. Turner, Mafalda Oliveira, Sacha J. Howell, Florence Dalenc, Javier Cortes, Henry L. Gomez Moreno, Xichun Hu, Komal Jhaveri, Petr Krivorotko, and Sibylle Loibl. Capivasertib in hormone receptor–positive advanced breast cancer. New England Journal of Medicine , 388(2...

    2023

  76. [84]

    Davies, Judit Debreczeni, Hannah Dry, Philippa Dudley, Ryan Greenwood, Paul D

    Matt Addie, Peter Ballard, David Buttar, Claire Crafter, Gordon Currie, Barry R. Davies, Judit Debreczeni, Hannah Dry, Philippa Dudley, Ryan Greenwood, Paul D. Johnson, Jason G. Kettle, Clare Lane, Gillian Lamont, Andrew Leach, Richard W. A. Luke, Jeff Morris, Donald Ogilvie, ...

    2013

  77. [85]

    Wu, Joshua Ballard, Brian B

    Kui Lin, Jie Lin, Wen-I. Wu, Joshua Ballard, Brian B. Lee, Susan L. Gloor, Guy P. A. Vigers, Tony H. Morales, Lori S. Friedman, and Nicholas Skelton. An ATP -site on-off switch that restricts phosphatase accessibility of AKT . Science Signaling , 5(223):ra37, 2012

    2012

  78. [87]

    Large language models think too fast to explore effectively

    Lan Pan, Hanbo Xie, and Robert C Wilson. Large language models think too fast to explore effectively. arXiv preprint arXiv:2501.18009 , 2025

    2025 arXiv

  79. [88]

    Evolve: Evaluating and optimizing llms forin-context exploration

    Allen Nie, Yi Su, Bo Chang, Jonathan Lee, Ed H Chi, Quoc V Le, and Minmin Chen. Evolve: Evaluating and optimizing llms forin-context exploration. In The Exploration in AI Today Workshop at ICML 2025 , 2025

    2025

  80. [89]

    The art and science of molecular docking

    Joseph M Paggi, Ayush Pandit, and Ron O Dror. The art and science of molecular docking. Annual review of biochemistry , 93(1):389--410, 2024

    2024

  81. [90]

    Accurate structure prediction of biomolecular interactions with alphafold 3

    Josh Abramson, Jonas Adler, Jack Dunger, Richard Evans, Tim Green, Alexander Pritzel, Olaf Ronneberger, Lindsay Willmore, Andrew J Ballard, Joshua Bambrick, et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature , 630(8016):493--500, 2024

    2024

  82. [91]

    Boltz-2: Towards accurate and efficient binding affinity prediction

    Saro Passaro, Gabriele Corso, Jeremy Wohlwend, Mateo Reveiz, Stephan Thaler, Vignesh Ram Somnath, Noah Getz, Tally Portnoi, Julien Roy, Hannes Stark, et al. Boltz-2: Towards accurate and efficient binding affinity prediction. BioRxiv , pages 2025--06, 2025

    2025

  83. [92]

    Claude 3.7 Sonnet and Claude code, February 2025

    Anthropic. Claude 3.7 Sonnet and Claude code, February 2025

    2025

  84. [93]

    OpenAI o3 and o4-mini system card, April 2025

    OpenAI. OpenAI o3 and o4-mini system card, April 2025

    2025

  85. [94]

    B., Yu, J., Soricut, R.,

    Team, G., Anil, R., Borgeaud, S., Alayrac, J. B., Yu, J., Soricut, R., ... & Blanco, L. (2023). Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805

    2023 arXiv

  86. [95]

    The virtual lab: Ai agents design new sars-cov-2 nanobodies with experimental validation

    Kyle Swanson, Wesley Wu, Nash L Bulaong, John E Pak, and James Zou. The virtual lab: Ai agents design new sars-cov-2 nanobodies with experimental validation. bioRxiv, pages 2024–11, 2024

    2024

  87. [96]

    Preuer, K., Renz, P., Unterthiner, T., Hochreiter, S., & Klambauer, G. (2018). Fréchet ChemNet distance: a metric for generative models for molecules in drug discovery. Journal of chemical information and modeling, 58(9), 1736-1741

    2018

  88. [97]

    Quantifying the chemical beauty of drugs

    G Richard Bickerton, Gaia V Paolini, Jérémy Besnard, Sorel Muresan, and Andrew L Hopkins. Quantifying the chemical beauty of drugs. Nature chemistry, 4(2):90–98, 2012

    2012

  89. [98]

    A., Lombardo, F., Dominy, B

    Lipinski, C. A., Lombardo, F., Dominy, B. W., & Feeney, P. J. (2012). Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Advanced drug delivery reviews, 64, 4-17

    2012

  90. [99]

    Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions

    Peter Ertl and Ansgar Schuffenhauer. Estimation of synthetic accessibility score of drug-like molecules based on molecular complexity and fragment contributions. Journal of cheminfor- matics, 1:1–11, 2009

    2009

  91. [100]

    A., & Crippen, G

    Wildman, S. A., & Crippen, G. M. (1999). Prediction of physicochemical parameters by atomic contributions. Journal of chemical information and computer sciences, 39(5), 868-873

    1999

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.