Pith. sign in

REVIEW 4 major objections 5 minor 32 references

CAi Copilot: Reducing Operational Workload in Molecular Design through Intent-Driven Agentic Workflows

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A three-layer intent-driven agent turns open-ended molecular design requests into traceable candidate-level evidence, scoring 84.59 on 45 curated tasks and beating five baselines.

desk verdict Useful agent system with a sensible architecture and credible external benchmarks, but the headline 18-point CAiMD lead rests on a run whose provenance is not yet auditable, and the workload claim is unmeasured. read the letter →

arxiv 2608.06961 v1 pith:KEPYDU3H submitted 2026-08-07 cs.AI cs.LG

Zhu Wang , Jiangyu Chen , Yingjun Shang , Yuhui Yao , Laiao Lu , Tianfan Fu , Na Zou This is my paper · ORCID
classification cs.AIcs.LG
keywords intent-to-evidencemoleculardesignworkflowAIagentthree-layerarchitectureCAiMDbenchmarkdrugdiscoveryevidencegroundingLLM
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

Early-stage molecular design is an iterative loop of proposal, assessment, filtering, ranking, and refinement, and this paper argues that the real bottleneck is not molecule generation but translating broad research intent into an executable, adaptive, traceable workflow. The central assertion is that CAi Copilot, a three-layer agent separating intent grounding, run-time control, and a molecular execution substrate, does this reliably: on 45 curated tasks it achieves an outcome score of 84.59, exceeding the next-best baseline by 18.07 points, with its largest edge in producing valid molecular outputs from tool calls. The authors also claim the benefit transfers to external benchmarks, SMDD-Bench, LIDDiA, and MolBench, on several subtasks under a shared backbone, while acknowledging that long-horizon integrated workflows and receptor-dependent docking remain weak. A sympathetic reader would care because the paper reframes the evaluation target: instead of scoring a molecule set, it scores a workflow record that connects each rank decision to candidate-level evidence an expert can inspect.

What carries the argument

The load-bearing object is the combined loop of three layers and a provenance-linked evidence record. The Research Interface Layer builds the initial plan $P_0 = (V_0, D_0)$, where $V_0$ is a set of steps $v_j = (g_j, u_j, x_j, p_j, e_j)$ carrying a local goal, a molecular operation, required inputs, input needs, and expected evidence, and $D_0$ records step dependencies. The Agent Reasoning Layer maintains the trajectory $h_t$ and evidence state $E_t$, selects $a_t \sim \pi(\cdot \mid h_t, P_t, E_t)$, executes it, updates $E_{t+1} = E_t \oplus \mathrm{Extract}(o_{t+1}, a_t)$, then computes the evidence gap $\Delta_{t+1} = \mathrm{Req}(\Gamma) \setminus \mathrm{Cov}(E_{t+1})$ and revises $P_t$ accordingly. The Execution Substrate gives every tool a contract $\kappa_k = (X_k, Y_k, \mathrm{Pre}_k, \mathrm{Exec}_k)$ that exposes accepted inputs, outputs, prerequisites, and invocation, and returns a standard observation with status, artifacts, warnings, and errors. The final output is $Y = (E, \tau, L)$ with $E = \{(m_i, z_i, r_i, \rho_i)\}$: candidate molecule, computed evidence, rank, and a provenance link $\rho_i$ to the supporting action, plus the trajectory $\tau$ and recorded limits $L$. This machinery carries the argument because the evidence gap is what makes plan revision reactive, and the provenance link is what makes every reported result traceable to a specific tool run.

What would settle it

Take the released CAiMD reference specifications and judge prompts, then have an independent panel of medicinal chemists re-score every agent's final reports with the agent identities removed; separately, have the panel author 45 fresh design requests that CAi's designers never see. If CAi's outcome-score lead over the strongest general-purpose baseline shrinks to statistical insignificance or reverses on the fresh requests, the performance claim is an artifact of the benchmark rather than a consequence of the three-layer architecture.

Watch

Extended reading notes

Core claim

The discovery the paper puts forward is architectural: reliable intent-to-evidence molecular design comes from separating three concerns, grounding the intent, controlling the run from interim results, and executing molecular computation through formal tool contracts, rather than from any single generator, planner, or tool. CAi's Research Interface Layer converts a design request into a plan of steps with explicit dependencies and expected evidence; the Agent Reasoning Layer executes each step, merges observations into candidate records, recomputes the remaining evidence gap $\Delta = \mathrm{Req}(\Gamma) \setminus \mathrm{Cov}(E)$, and updates the plan; and the Execution Substrate runs molecular operations behind uniform contracts that expose inputs, outputs, prerequisites, and invocation, returning standardized observations. The paper's evidence is the 45-task CAiMD comparison, where CAi reaches 84.59 on objective satisfaction versus 66.52 for the strongest baseline, alongside external benchmarks showing transfer on executable screening, editing, and optimization while exposing limits in long-horizon execution and structure-dependent workflows. The target-specific JAK1/JAK2 case study is offered as a demonstration that the workflow prioritizes the requested predicted-affinity objective (median Vina score of -8.706 kcal/mol, with 18 of 24 selected candidates below the -8.0 threshold) while preserving explicit evidence on selectivity and property trade-offs.

Load-bearing premise

The load-bearing premise is that the 45-task CAiMD benchmark, whose reference specifications were manually authored by the same team that built CAi and whose semantic scores come from a blinded LLM judge with prompts slated for later release, validly represents real expert molecular-design workflows and is not biased toward CAi's reporting style; if that premise fails, the 18.07-point outcome lead over the next-best agent would not transfer to uncurated expert requests.

Editorial extensions

If this is right

  • On the 45 CAiMD tasks, CAi's valid-tool-output rate of 67.49% versus 34.91% for the strongest baseline implies that the execution stage, turning planned operations into usable molecular artifacts, is the main bottleneck in agentic molecular design, not intent understanding or task decomposition.
  • With a shared backbone, CAi's SMDD-Bench success rates improve from 20.0% to 24.0% on 2D Pharmacophore Identification and from 50.0% to 63.3% on Lead Optimization, indicating that the same workflow control transfers to independent task evaluators.
  • On LIDDiA's 30 targets, CAi fills 39.2 of 50 candidate slots with valid molecules versus 14.4 for the reference agent (78.5% versus 28.7% valid), suggesting its advantage is broad candidate coverage and constraint satisfaction rather than uniform per-molecule quality.
  • On MolBench's 190 examples, CAi raises MS-1 screening accuracy from 0.1800 to 0.9600 and overall editing correctness from 0.8718 to 0.9744, while the reference agent remains stronger on receptor-dependent virtual screening and scaffold preservation, a trade-off the paper attributes to limited cross-environment structural visibility.
  • Because strict success falls most sharply on integrated workflows while objective satisfaction stays stable, CAi accumulates grounded partial evidence even in long trajectories but does not yet complete every dependency; long-horizon execution is the paper's own stated remaining limit.
  • The target-specific case shows CAi's output is a selected portfolio rather than raw generator outputs, so the threshold-pass comparison (18 of 24 versus 7 of 80) mainly demonstrates workflow-level prioritization and leaves experimental activity outside the agent's claims.

Reading between the lines

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

  • The 45 CAiMD reference specifications were authored by the same team that built CAi and are scored partly by a blinded LLM judge whose prompts are slated for later release; until an independent panel re-scores the runs blind to agent identity, the 18.07-point outcome gap should be read as benchmark-relative rather than a universal capability claim.
  • The evidence-gap control loop $\Delta = \mathrm{Req}(\Gamma) \setminus \mathrm{Cov}(E)$ is a generic primitive, so a natural test is to port CAi's three-layer separation to other closed-loop scientific workflows, such as materials optimization or retrosynthesis, and check whether the same execution-stage gains appear where interim observations change later steps.
  • The paper's own diagnosis implies a cheap ablation: freeze the plan after the Research Interface Layer so the Agent Reasoning Layer cannot revise it, and measure the drop in valid-tool-output rate; if the drop is small, most of CAi's gain comes from tool contracts and the execution substrate rather than from adaptive reasoning.
  • The scaffold-based case comparison of 18/24 threshold-passing candidates versus 7/80 raw baseline molecules reflects workflow-level prioritization, not controlled generator superiority; a testable extension is whether that selected portfolio improves an external oracle's true-positive rate in a prospective docking or assay setting.
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 / 5 minor

Summary. The paper proposes CAi Copilot, a three-layer agentic workflow system for early-stage molecular design, formulated as intent-to-evidence execution. The Research Interface Layer converts research intent into an executable plan, the Agent Reasoning Layer adapts the plan from interim results, and the Execution Substrate provides molecular tools and metrics. The authors evaluate CAi on their own 45-task CAiMD benchmark against five baselines, reporting an outcome score of 84.59 versus 66.52 for the next-best system, and on external benchmarks (SMDD-Bench, LIDDiA, MolBench) with native protocols. They conclude that reliable molecular-design execution requires joint strength across planning, execution, and evidence grounding, with long-horizon integrated workflows remaining a limitation.

Significance. If the claims hold, the paper offers a useful task formulation and a modular architecture that makes workflow planning, execution, and evidence tracing explicit. The external benchmark results (Tables 3-5) provide partial independent validation, and the paper is transparent about several limitations, including the exclusion of efficiency measurements and the absence of experimental validation. However, the headline CAiMD comparison rests on a self-authored benchmark whose semantic rubric is not yet released, and an appendix statement suggests CAi's trajectory was a pre-existing reference run rather than a live baseline-supervisor run, which could undermine the main head-to-head result. The 'operational workload reduction' in the title is also never directly measured. The core architectural ideas are promising, but the main quantitative claims require clarification and re-analysis before the paper can be accepted.

major comments (4)
  1. [§4.1 and Appendix B.2] Section 4.1 states that for CAiMD 'all agents use the same DeepSeek V4 Pro-compatible endpoint, prompts, inputs, isolated workspaces, and 2,400s budget.' Appendix B.2 states that CAi is 'a previously completed reference run normalized and scored under the same evaluation rubric, rather than a sixth queue in the baseline supervisor.' These statements are not obviously compatible. If CAi's trajectory was produced before the live baseline harness existed, the 2,400s budget, the no-retry policy, the prompt template, and the artifact-contract adapter may not have been enforced for CAi. Because the headline outcome lead (84.59 vs. 66.52 in Tables 1 and 2) comes exclusively from CAiMD, this asymmetry is load-bearing. Please clarify whether CAi was run under the identical live conditions as the baselines, and if it was not, provide a rerun under those conditions or explicitly present the CAiMD head-to-head as a pilot rather than the primary comparison.
  2. [§4.1, Appendix A.3, Appendix B.1] CAiMD is a benchmark whose 45 reference specifications were manually authored by the same team that designed CAi, and the semantic scoring relies on a blinded LLM judge with rubrics that are not included in the paper: Appendix B.1 says 'The final rubric and judge prompts will be released with the evaluation suite.' The rule-based metrics in Eqs. (10)-(13) are transparent, but the judge-dependent components (task decomposition, semantic trajectory precision, hallucination-free interpretation, and others) are not auditable. Because the benchmark is self-authored and the judge prompts are unavailable, the 18.07-point lead over Codex could partly reflect rubric or design alignment, or judge bias favoring CAi's reporting style. Please release the full benchmark specifications and judge prompts, report judge agreement, and provide evidence that the reference specifications were not shaped by CAi's design choices; otherwise the CAiMD comparison should be treated as a self-assessment rather than the primary evidence of superiority.
  3. [§4.3, Table 2 note; title] The paper's title and abstract claim 'reducing operational workload,' and the introduction frames the contribution in terms of researchers needing to coordinate steps and integrate evidence. However, Table 2 states: 'Efficiency is excluded because matched manual-time measurements are not yet available.' No measurement of workload, wall-clock time, human effort, number of interventions, or cost appears anywhere in the manuscript. Thus the central stated benefit is not empirically supported. Please either add workload or efficiency measurements on at least a subset of tasks, or revise the title and abstract to claim 'workflow reliability' or 'evidence grounding' rather than workload reduction.
  4. [Tables 1, 3, 4, 5] The results are reported as single comparisons without confidence intervals, significance tests, or multiple seeds. For example, the SMDD improvements in Table 3 (20.0% to 24.0% on 25 instances; 0.0% to 1.7% on 60 instances) correspond to one additional solved instance per task type, and the MolBench results in Table 5 are reported without variance. Given that the headline CAiMD lead is already sensitive to run-condition asymmetries, the absence of uncertainty quantification makes it difficult to assess whether the observed gaps are stable. Please report multiple seeds or bootstrapped confidence intervals for at least the main CAiMD and external benchmark comparisons, or explicitly state the number of runs and the variance.
minor comments (5)
  1. [Appendix B.1] The notation TA_i and TP_i in Eq. (10) is described as 'rule-based trajectory accuracy and trajectory precision proxies,' but Table 1 groups TA and TP under 'Workflow' without defining them at first use in the main text; please align the notation and define all nine metric abbreviations when they are introduced.
  2. [References [8], [10], [30]] Several references contain the placeholder text 'Please verify full author list before final submission'; this must be resolved before publication.
  3. [Figure 2] The figure caption and in-text citations use labels (a) through (e), but the figure image appears to repeat panel labels A and B; please renumber the panels consistently with the text.
  4. [§4.1 first paragraph] The phrase 'all agents use the same ... prompts' should specify that this refers to task inputs rather than system prompts, since each baseline retains its native control logic and the adapters in Appendix B.2 necessarily use different system-level instructions.
  5. [§4.2 and Table 1] All systems score exactly 95.56% on intent understanding, which suggests a ceiling effect; please comment on whether this metric is too coarse to differentiate intent-recovery capability, or whether the near-constant score is an artifact of the scoring definition.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central evaluation is an empirical comparison, and the independent external benchmarks corroborate the agent claims.

full rationale

No circular step is present. The paper's derivation chain is: formulate intent-to-evidence workflow execution, build the three-layer CAi Copilot, then measure it on CAiMD and on external benchmarks with released evaluators. The CAiMD reference specifications are manually curated from published molecular-design studies and PDB-derived inputs, not from CAi's outputs; the metrics in Equations (10)-(13) are set-based ratios comparing reference operation sets, invocation sets, claim sets, and objective sets against the agent's executed trajectory. Nothing in those equations forces a particular agent's score to equal its own inputs, so the headline result (84.59 versus 66.52 on objective satisfaction) is an empirical measurement rather than a tautology. Appendix B.2 states that 'CAi is a previously completed reference run normalized and scored under the same evaluation rubric, rather than a sixth queue in the baseline supervisor.' That raises a legitimate experimental-comparability concern about whether run conditions matched the live baseline queues, but it is not circularity: a potentially unmatched run is a validity threat, not a definitional reduction of the claimed outcome to the input. The paper also reports results on SMDD-Bench, LIDDiA, and MolBench using native protocols and released evaluators, which provide independent evidence independent of the CAiMD rubric. There are no load-bearing self-citations: the reference list contains no overlapping-author self-citation that is invoked to justify the central claim. The statement that 'The final rubric and judge prompts will be released with the evaluation suite' is a transparency limitation, but unavailability of a scoring artifact does not by itself make the reasoning circular. Overall, the central derivation is self-contained and externally anchored.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

No free model parameters are fitted; the central claim depends instead on hand-chosen benchmark thresholds and on assumptions that docking scores, LLM judging, and the curated task set are valid measurements of molecular-design workflow quality. The external benchmarks provide some independent grounding, but the main CAiMD instrument is not yet public.

free parameters (1)
  • CAiMD task thresholds and metric directions = varies per task
    Hand-specified in the reference specifications (Appendix A.3); they define objective satisfaction and thus directly set the headline OS scores.
assumptions (4)
  • ad hoc to paper The 45 CAiMD tasks and their reference specifications are representative and unbiased proxies for expert molecular design workflows.
    Tasks are manually curated by the authors (Appendix A); no external validation or blind expert review of the reference specs is reported.
  • domain assumption Vina docking scores and the released evaluators (RDKit, ADMET-AI, Boltz2) are treated as valid candidate-level evidence.
    Used throughout Sections 4.6 and Appendix D to define affinity and property outcomes; no experimental verification is claimed.
  • domain assumption A blinded LLM judge scores semantic criteria (task decomposition, semantic trajectory precision, hallucination-free interpretation) reliably and without favoring CAi.
    Appendix B.1 describes the judge but reports no inter-rater agreement, calibration, or validation against human experts.
  • domain assumption Trajectory normalization across heterogeneous agent systems is faithful, so capability-level comparisons are fair.
    Appendix B.2 maps native trajectories to capability families; this mapping is authored by the evaluators and could encode assumptions that favor certain systems.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CAi Copilot: Reducing Operational Workload in Molecular Design through Intent-Driven Agentic Workflows." pith.science (2026). https://pith.science/paper/KEPYDU3H

@misc{pith2026260806961,
  author       = {Pith},
  title        = {Pith review of: CAi Copilot: Reducing Operational Workload in Molecular Design through Intent-Driven Agentic Workflows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KEPYDU3H}},
  note         = {Machine review of arXiv:2608.06961}
}
read the original abstract

Early-stage molecular design is an iterative process, not just a task of generating molecules. Researchers turn broad goals into design strategies, refine candidates, assess many properties, and gather evidence before synthesis and tests. AI methods can generate molecules, optimize several goals, predict properties, dock compounds, and account for synthesis. Yet these functions are spread across specialized tools. Experts must still coordinate each step, judge interim results, and integrate evidence. The central challenge is thus to turn research intent into adaptive, traceable runs grounded in scientific tools. We cast this challenge as intent-to-evidence molecular design workflow execution and present CAi Copilot, an expert-oriented agent with three linked layers. The Research Interface Layer turns intent into an executable plan. The Agent Reasoning Layer uses interim results to guide each run. The Execution Substrate supplies molecular tools, metrics, reusable utilities, and backend services. Across 45 tasks, CAi achieves the strongest overall performance, with an outcome score of 84.59, exceeding the next-best result by 18.07 points. Additional benchmarks test how CAi coordinates generation, screening, and multi-criteria evaluation, while exposing limits in long-horizon execution. These results show that CAi turns broad molecular-design intent into transparent, traceable workflows that connect interim decisions to candidate-level evidence.

Figures

Figures reproduced from arXiv: 2608.06961 by the authors.

Figure 1
Figure 1. Overview of CAi Copilot. The three coordinated layers translate research intent into an [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Workflow reachability and objective-conditioned molecular evidence in the JAK-family [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Scaffold-based distributions of predicted JAK1 Vina score, docking-based ligand efficiency, [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Physicochemical distributions for workflow-matched scaffold-based outputs. Molecular [PITH_FULL_IMAGE:figures/full_fig_p019_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 13 canonical work pages

  1. [1]

    Baker, Ian A

    Reza Averly, Frazier N. Baker, Ian A. Watson, and Xia Ning. Liddia: Language-based intelligent drug discovery agent. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 12004–12028, 2025. doi: 10.18653/v1/2025.emnlp-main.603. URL https://aclanthology.org/2025.emnlp-main.603/

    doi:10.18653/v1/2025.emnlp-main.603 2025

  2. [2]

    An ai system to help scientists write expert-level empirical software

    Eser Aygün et al. An ai system to help scientists write expert-level empirical software. Nature, 654:909–916, 2026. doi: 10.1038/s41586-026-10658-6. URLhttps://www.nature.com/ articles/s41586-026-10658-6

    doi:10.1038/s41586-026-10658-6 2026

  3. [3]

    Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D

    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, 6:525–535, 2024. doi: 10.1038/s42256-024-00832-8. URLhttps://www.nature.com/articles/ s42256-024-00832-8

    doi:10.1038/s42256-024-00832-8 2024

  4. [4]

    Synflownet: Design of diverse and novel molecules with synthesis constraints.arXiv preprint arXiv:2405.01155, 2024

    Miruna Cretu, Charles Harris, Ilia Igashov, Arne Schneuing, Marwin Segler, Bruno Correia, Julien Roy, Emmanuel Bengio, and Pietro Liò. Synflownet: Design of diverse and novel molecules with synthesis constraints.arXiv preprint arXiv:2405.01155, 2024. doi: 10.48550/ arXiv.2405.01155. URLhttps://arxiv.org/abs/2405.01155

    arXiv doi:10.48550/arxiv.2405.01155 2024

  5. [5]

    Macrocyclization of linear molecules by deep 12 learning to facilitate macrocyclic drug candidates discovery.Nature Communications, 14: 4552, 2023

    Yanyan Diao, Dandan Liu, Huan Ge, Rongrong Zhang, Kexin Jiang, Runhui Bao, Xiaoqian Zhu, Hongjie Bi, Wenjie Liao, Ziqi Chen, et al. Macrocyclization of linear molecules by deep 12 learning to facilitate macrocyclic drug candidates discovery.Nature Communications, 14: 4552, 2023. doi: 10.1038/s41467-023-40219-8. URL https://www.nature.com/articles/ s41467-...

    work page doi:10.1038/s41467-023-40219-8 2023

  6. [6]

    Machine learning-aided generative molecular design.Nature Machine Intelligence, 6:589–604, 2024

    Yuanqi Du et al. Machine learning-aided generative molecular design.Nature Machine Intelligence, 6:589–604, 2024. doi: 10.1038/s42256-024-00843-5. URLhttps://www.nature. com/articles/s42256-024-00843-5

    doi:10.1038/s42256-024-00843-5 2024

  7. [7]

    Translation between molecules and natural language

    Carl Edwards, Tuan Lai, Kevin Ros, Garrett Honke, Kyunghyun Cho, and Heng Ji. Translation between molecules and natural language. InProceedings of the 2022 Con- ference on Empirical Methods in Natural Language Processing, pages 375–413, 2022. doi: 10.18653/v1/2022.emnlp-main.26. URL https://aclanthology.org/2022.emnlp-main.26/

    doi:10.18653/v1/2022.emnlp-main.26 2022

  8. [8]

    Mol-instructions: A large-scale biomolecular instruction dataset for large language models.arXiv preprint arXiv:2306.08018, 2023

    Yin Fang et al. Mol-instructions: A large-scale biomolecular instruction dataset for large language models.arXiv preprint arXiv:2306.08018, 2023. URL https://arxiv.org/abs/ 2306.08018. Please verify full author list before final submission

    arXiv 2023

Show all 32 references
  1. [9]

    Generation of 3d molecules in pockets via language model.Nature Machine Intelligence, 6:62–73, 2024

    Wei Feng, Lvwei Wang, Zaiyun Lin, Yanhao Zhu, Han Wang, Jianqiang Dong, Rong Bai, Huting Wang, Jielong Zhou, Wei Peng, Bo Huang, and Wenbiao Zhou. Generation of 3d molecules in pockets via language model.Nature Machine Intelligence, 6:62–73, 2024. doi: 10. 1038/s42256-023-0077...

    2024

  2. [10]

    Txagent: An ai agent for therapeutic reasoning across a universe of tools

    Shanghua Gao et al. Txagent: An ai agent for therapeutic reasoning across a universe of tools. arXiv preprint arXiv:2503.10970, 2025. URL https://arxiv.org/abs/2503.10970. Please verify full author list before final submission

    2025 arXiv

  3. [11]

    Ghareeb et al

    Ali E. Ghareeb et al. A multi-agent system for automating scientific discovery.Nature, 655: 497–505, 2026. doi: 10.1038/s41586-026-10652-y. URL https://www.nature.com/articles/ s41586-026-10652-y

    2026 doi

  4. [12]

    Accelerating scientific discovery with co-scientist.Nature, 655:487– 496, 2026

    Juraj Gottweis et al. Accelerating scientific discovery with co-scientist.Nature, 655:487– 496, 2026. doi: 10.1038/s41586-026-10644-y. URL https://www.nature.com/articles/ s41586-026-10644-y

    2026 doi

  5. [13]

    Decompdiff: Diffusion models with decomposed priors for structure- based drug design.Proceedings of the 40th International Conference on Machine Learning, 202: 11827–11846, 2023

    Jiaqi Guan, Xiangxin Zhou, Yuwei Yang, Yu Bao, Jian Peng, Jianzhu Ma, Qiang Liu, Liang Wang, and Quanquan Gu. Decompdiff: Diffusion models with decomposed priors for structure- based drug design.Proceedings of the 40th International Conference on Machine Learning, 202: 11827–1...

    2023

  6. [14]

    Improving de novo molecular design with curriculum learning.Nature Machine Intelligence, 4:555–563, 2022

    Jeff Guo, Vendy Fialková, Juan Diego Arango, Christian Margreitter, Jon Paul Janet, Kostas Papadopoulos, Ola Engkvist, and Atanas Patronov. Improving de novo molecular design with curriculum learning.Nature Machine Intelligence, 4:555–563, 2022. doi: 10.1038/ s42256-022-00494-...

    2022

  7. [15]

    SMDD-Bench: Can LLMs solve real-world small molecule drug design tasks?, 2026

    Kevin Han, Renfei Zhang, Kathy Wei, Hamed Mahdavi, Niloofar Mireshghallah, and Amir Farimani. SMDD-Bench: Can LLMs solve real-world small molecule drug design tasks?, 2026. URLhttps://arxiv.org/abs/2605.21740

    2026 arXiv

  8. [16]

    Exploring the macrocyclic chemical space for heuristic drug design with deep learning models.Communications Chemistry, 8:299, 2025

    Feng Hu, Xiaotong Jia, Wenjie Liao, Ziqi Chen, Hongjie Bi, Huan Ge, Dandan Liu, Rongrong Zhang, et al. Exploring the macrocyclic chemical space for heuristic drug design with deep learning models.Communications Chemistry, 8:299, 2025. doi: 10.1038/s42004-025-01686-w. URLhttps:...

    2025 doi

  9. [17]

    Carter, Xin Zhou, Matthew Wheeler, Jonathan A

    Kexin Huang, Serena Zhang, Hanchen Wang, Yuanhao Qu, Yingzhou Lu, Yusuf Roohani, Ryan Li, Lin Qiu, Gavin Li, Junze Zhang, Di Yin, Shruti Marwaha, Jennefer N. Carter, Xin Zhou, Matthew Wheeler, Jonathan A. Bernstein, Mengdi Wang, Peng He, Jingtian Zhou, Michael Snyder, Le Cong,...

    2025 doi

  10. [18]

    Michał Koziarski, Andrei Rekesh, Dmytro Shevchuk, Almer van der Sloot, Piotr Gaiński, Yoshua Bengio, Cheng-Hao Liu, Mike Tyers, and Robert A. Batey. Rgfn: Synthesizable molecular generation using gflownets.arXiv preprint arXiv:2406.08506, 2024. doi: 10.48550/arXiv.2406. 08506....

    2024 doi arXiv

  11. [19]

    Druga- gent: Automating ai-aided drug discovery programming through llm multi-agent collaboration

    Sizhe Liu, Yizhou Lu, Siyu Chen, Xiyang Hu, Jieyu Zhao, Yingzhou Lu, and Yue Zhao. Druga- gent: Automating ai-aided drug discovery programming through llm multi-agent collaboration. arXiv preprint arXiv:2411.15692, 2024. URLhttps://arxiv.org/abs/2411.15692

    2024 arXiv

  12. [20]

    Loeffler, Jiazhen He, Alessandro Tibo, Jon Paul Janet, Anton Voronov, Lewis H

    Hannes H. Loeffler, Jiazhen He, Alessandro Tibo, Jon Paul Janet, Anton Voronov, Lewis H. Mervin, and Ola Engkvist. Reinvent 4: Modern ai-driven generative molecule design.Journal of Cheminformatics, 16(1):20, 2024. doi: 10.1186/s13321-024-00812-5. URLhttps://jcheminf. biomedce...

    2024 doi

  13. [21]

    Hermes agent

    Nous Research. Hermes agent. Open-source agent software, https://github.com/ NousResearch/hermes-agent, 2026. Evaluated software version and commit to be reported

    2026

  14. [22]

    Aluru, Achuth Chandrasekhar, and Amir Barati Farimani

    Janghoon Ock, Radheesh Sharma Meda, Srivathsan Badrinarayanan, Neha S. Aluru, Achuth Chandrasekhar, and Amir Barati Farimani. Large language model agent for modular task execution in drug discovery.Journal of Chemical Information and Modeling, 66(4):2055–2068,

  15. [23]

    Addendum to openai o3 and o4-mini system card: Codex

    OpenAI. Addendum to openai o3 and o4-mini system card: Codex. Technical report, OpenAI, May 2025. URLhttps://cdn.openai.com/pdf/8df7697b-c1b2-4222-be00-1fd3298f351d/ codex_system_card.pdf

    2025

  16. [24]

    Pocket2mol: Efficient molecular sampling based on 3d protein pockets.International Conference on Machine Learning, 2022

    Xingang Peng, Shitong Luo, Jiaqi Guan, Qi Xie, Jian Peng, and Jianzhu Ma. Pocket2mol: Efficient molecular sampling based on 3d protein pockets.International Conference on Machine Learning, 2022. doi: 10.48550/arXiv.2205.07249. URLhttps://arxiv.org/abs/2205.07249

    2022 doi arXiv

  17. [25]

    Blum, and Lars Ruddigkeit

    Jean-Louis Reymond, Ruud van Deursen, Lorenz C. Blum, and Lars Ruddigkeit. Chemical space as a source for new drugs.MedChemComm, 1(1):30–38, 2010. doi: 10.1039/C0MD00020E. URLhttps://pubs.rsc.org/en/content/articlehtml/2010/md/c0md00020e

    2010 doi

  18. [26]

    Computer-based de novo design of drug-like molecules

    Gisbert Schneider and Uli Fechner. Computer-based de novo design of drug-like molecules. Nature Reviews Drug Discovery, 4(8):649–663, 2005. doi: 10.1038/nrd1799. URL https: //www.nature.com/articles/nrd1799

    2005 doi

  19. [27]

    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. doi: 10.1002/jcc.21334. URL https: //pubmed.ncbi.nlm.nih....

    2010

  20. [28]

    Diffdec: Structure-aware scaffold decoration with an end-to-end diffusion model.Journal of Chemical Information and Modeling, 14 64(7):2554–2564, 2024

    Junjie Xie, Sheng Chen, Jinping Lei, and Yuedong Yang. Diffdec: Structure-aware scaffold decoration with an end-to-end diffusion model.Journal of Chemical Information and Modeling, 14 64(7):2554–2564, 2024. doi: 10.1021/acs.jcim.3c01466. URLhttps://pubs.acs.org/doi/10. 1021/ac...

    2024 doi

  21. [29]

    Baker, Ziqi Chen, Xia Ning, and Huan Sun

    Botao Yu, Frazier N. Baker, Ziqi Chen, Xia Ning, and Huan Sun. Llasmol: Advancing large language models for chemistry with a large-scale, comprehensive, high-quality instruction tuning dataset. InFirst Conference on Language Modeling, 2024. URL https://openreview.net/ forum?id...

    2024

  22. [31]

    Molclaw: An autonomous agent with hierarchical skills for drug molecule evaluation, screening, and optimization.arXiv preprint arXiv:2604.21937, 2026

    Lisheng Zhang, Lilong Wang, Xiangyu Sun, Wei Tang, Haoyang Su, Yuehui Qian, Qikui Yang, Qingsong Li, Zhenyu Tang, Haoran Sun, Yingnan Han, Yankai Jiang, Wenjie Lou, Bowen Zhou, Xiaosong Wang, Lei Bai, and Zhengwei Xie. Molclaw: An autonomous agent with hierarchical skills for ...

    2026 doi arXiv

  23. [2024]

    Please verify full author list before final submission

    URL https://arxiv.org/abs/2402.06852. Please verify full author list before final submission

    arXiv

  24. [2026]

    URLhttps://doi.org/10.1021/acs.jcim.5c02454

    doi: 10.1021/acs.jcim.5c02454. URLhttps://doi.org/10.1021/acs.jcim.5c02454

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.