DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery

Yuxing Chen; Zhisheng Zhang; Cheng Tang; Wei Li; Minghui Zhang; Ziyan Huang; Shaocong Wei; Jiawei Chen

cs.AIOct 16, 2026

DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery

Authors: Yuxing Chen, Zhisheng Zhang, Cheng Tang, Wei Li, Minghui Zhang, Ziyan Huang, Shaocong Wei, Jiawei Chen,

Organizations: Shanghai Artificial Intelligence Laboratory · China Pharmaceutical University · University of the Chinese Academy of Sciences · Shanghai Jiao Tong University ·

Abstract

Computational drug discovery requires agents to compose scientific operations and adapt to changing evidence and constraints. Here we introduce DrugHarness, an evolvable agent harness coupling composable scientific execution with conditional skill induction. Its environment integrates over 80 scientific operations, biomedical resources and task-specific code, preserving associations among candidates, results and artifacts. From execution trajectories and outcome feedback, DrugHarness induces action hypotheses with observable applicability conditions, recommended actions and supporting evidence. An independent review agent evaluates candidate skills derived from these hypotheses through source verification and, when needed, targeted execution trials. Admitted skills guide subsequent tasks and are revised through feedback. Evaluation on MolBench and SMDD-Bench demonstrates capabilities in molecular screening, optimization and constrained design. Without evolution, DrugHarness exceeds reported MolClaw-CC results by 10.8 percentage points in affinity-comparison accuracy and 55% in mean top-three active hits on MolBench. Retrospective evaluation on skill-development tasks shows that evolution increases Qwen's affinity-comparison accuracy from 62.2% to 81.1% and optimization success from 79.5% to 87.2%. In a prospective HPK1 study, DrugHarness achieves a more favorable median docking score than Biomni among system-specific candidates. A synthesized compound achieves a cellular IC50 of 14.19 nM, a 2.64-fold potency improvement over the best input reference, with activity competitive with two selected compounds designed by experienced medicinal chemists. These results support conditional skill evolution for scientific agents and demonstrate a workflow connecting computational drug design with experimental validation.

Figures & tables

Figure 1: Overview of DrugHarness. (A) The agent composes public evidence, scientific tools, and code into task-specific workflows around shared scientific objects. Traces and permitted outcome feedback form experiences. (B) The learner induces conditional action hypotheses and translates them into candidate skills. An independent reviewer checks their evidence and conducts targeted trials in the same workspace when needed. Admitted skills guide subsequent tasks, while review records update insights and history for further learning.

Method / Runtime	Model	2D Pharm. $\uparrow$	IP $\uparrow$	SH $\uparrow$	LO $\uparrow$	FA $\uparrow$
SMDD harness	Claude Sonnet 4.6	28.0	0.0	3.8	47.5	0.0
	GPT-5.4	12.0	0.0	3.8	55.0	1.7
	Gemini 3.1 Pro	20.0	4.0	0.0	47.5	1.7
	Kimi K2.5 Thinking	12.0	0.0	1.9	32.5	0.0
	DeepSeek V3.2	8.0	0.0	3.8	22.5	0.0
	MiniMax M2.7	16.0	0.0	1.9	27.5	0.0

Table 1: Results on SMDD-Bench. Success rates (%) on the same task set, including a fixed subset of 40 lead-optimization tasks.

Method	MolClaw comparison group			CAi	DrugHarness (w/o Evol.)
Model	Sonnet 4.6	Sonnet 4.6	Qwen3.7- Max	Qwen3.7- Max	Qwen3.7- Max	GPT5.6- Terra	GPT5.6- Terra
Runtime	Claude Code †	Claude Code	N/R	CAi	Codex	Hermes	Codex
Screening
Accuracy (%) $\uparrow$	98.0	98.0	18.0	96.0	96.0	98.0	98.0
Precision (%) $\uparrow$	—	—	32.0	98.0	98.0	98.0	100.0
Recall (%) $\uparrow$	—	—	44.5	97.0	97.0	98.0	99.0

Table 2: Results on MolBench. Baseline results are from Zhang et al. (2026b) and Wang et al. (2026) . Metric definitions are provided in Appendix A.1 .

Setting	Screening	Affinity	Virtual screening			Editing	Optimization
	Acc. $\uparrow$	Acc. $\uparrow$	Hit@3 $\uparrow$	Mean Hits@3 $\uparrow$	Mean GT rank $\downarrow$	Acc. $\uparrow$	SR $\uparrow$	Valid JSON $\uparrow$
w/o Evol.	96.0	62.2	60.0	0.72	29.05	97.4	79.5	87.2
Evol.	100.0	81.1	48.0	0.64	26.01	97.4	87.2	100.0

Table 3: Experience-driven evolution on MolBench. Codex/Qwen3.7-Max with historical no-skill runs and retrospective in-domain evaluation using frozen domain-specific skills. Proportions are percentages. Complete metrics: Appendix A.2 .

Panel of Figure 2 — Figure 2: Computational evaluation of HPK1 candidates. Four system-specific structures per group remain after excluding six shared structures from ten evaluated entries each. Colors: DrugHarness; Biomni. Symbols: molecules ( $\bullet$ ), medians ( ). Lower is favorable. The four inhibition endpoints were selected post hoc for lower DrugHarness medians; its median hERG inhibition was higher (0.949 versus 0.663) and QED lower (0.3645 versus 0.4085).

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Property	Tasks	Valid pairs	Positive gain	Mean gain	Scaffold retained
LogP	11	10	10	0.9260	4/11
QED	20	19	19	0.3748	3/20
Solubility	8	8	8	0.7089	7/8

Appendix

Table 4: Property optimization with Codex/Terra. Valid pairs denote valid scoring pairs; scaffold retention uses the hard criterion.

Metric	w/o Evol.	Evol.
Screening
Accuracy (%) $\uparrow$	96.0	100.0
Precision (%) $\uparrow$	98.0	100.0
Recall (%) $\uparrow$	97.0	100.0
F1 (%) $\uparrow$	97.3	100.0
Specificity (%) $\uparrow$	100.0	100.0

Appendix

Table 5: Complete MolBench evolution metrics. Both configurations use Codex/Qwen3.7-Max. Proportions are percentages; the mean

\Delta

is the descriptive task-wise mean of saved native improvements.

Capability	Tools and resources	Inputs $\rightarrow$ outputs
Biomedical evidence retrieval	Open Targets ( Buniello et al., 2025 ) ; cBioPortal ( Cerami et al., 2012 ; Gao et al., 2013 ) ; TTD ( Zhou et al., 2024 ) ; ClinVar ( Landrum et al., 2025 ) ; GTEx ( The GTEx Consortium, 2020 ) ; ENCODE ( The ENCODE Project Consortium, 2020 ) ; STRING ( Szklarczyk et al., 2023 ) ; Europe PMC ( Rosonovski et al., 2024 ) ; ClinicalTrials.gov and FDA record interfaces.	Disease, target, variant, or experimental context $\rightarrow$ associations, variants, expression, functional and network evidence, publications, and study records with provenance.
Target identity and structural data	UniProt ( The UniProt Consortium, 2025 ) ; RCSB PDB ( Burley et al., 2022 ) ; SIFTS ( Velankar et al., 2013 ) ; AlphaFold DB ( Varadi et al., 2024 ) ; NAKB (nucleic-acid structures) ( Lawson et al., 2024 ) ; KinHub ( 28 ) .	Names, sequences, or database identifiers $\rightarrow$ target identities, sequences, structure candidates and coordinates, chain/residue mappings, and family annotations.
Receptor, ligand, and pocket preparation	PDBFixer ( PDBFixer contributors, ) ; Meeko ( Santos-Martins et al., 2025 ) ; RDKit ( RDKit contributors, ) ; CCD ( Westbrook et al., 2015 ) ; fpocket ( Le Guilloux et al., 2009 ) ; P2Rank ( Krivák and Hoksza, 2018 ) .	Protein/ligand structures, component selections, and preparation parameters $\rightarrow$ prepared receptors and ligand states, pocket candidates, docking grids, and input files.
Interaction and pharmacophore analysis	PLIP ( Schake et al., 2025 ) ; ProLIF ( Bouysset and Fiorucci, 2021 ) ; RDKit; atomic geometry analysis and co-crystal alignment/clustering.	Complexes, structural ensembles, or labeled molecules $\rightarrow$ contacts, water-mediated paths, interaction fingerprints, conserved three-dimensional interaction points, and molecular features.
Compound sources and activity evidence	ChEMBL ( Zdrazil et al., 2024 ) ; BindingDB ( Liu et al., 2025 ) ; PubChem ( Kim et al., 2025 ) ; local-library similarity search; Google Patents retrieval by publication number.	Target/compound identifiers, reference molecules, or patent publication numbers $\rightarrow$ chemical identities, measured activities with assay context, candidate rankings, or patent documents.
Molecule generation and fragment linking	DiffSBDD ( Schneuing et al., 2024 ) ; FlowMol3 ( Dunn and Koes, 2026 ) ; AutoFragDiff ( Ghorbani et al., 2024 ) ; explicit molecular-graph linking.	Generation settings, optional pocket context, or fragments with attachment sites $\rightarrow$ three-dimensional candidates, fragment mappings, and linked-molecule proposals.

Appendix

Table 6: Scientific capabilities of DrugHarness. Representative tools and resources supporting over 80 agent-accessible scientific operations are organized into ten capability categories. A tool may support multiple operations; entries summarize capabilities rather than enumerate individual interfaces.

Figure 4: Reference and generated BACE1 candidates. B2-09 and B3-02 were both generated from the reference molecule. Structures are aligned by their shared core.

Table 7: ADMET-AI predictions for the BACE1 candidates. B2-09 meets three of the four improvement targets but exceeds the CYP3A4 limit. B3-02 meets all four targets, and both candidates satisfy the four holding conditions.

Figure 5: Candidate-linked assessment and submission. The three assessments refer to the same registered B3-02 molecule. Following assessment, the candidate was written to solution.smi , read back and submitted in the evaluated molecular state.

Panel of Figure 6 — Figure 6: Predicted property trade-offs among BACE1 candidates. (a) All 32 evaluated structures, including the reference; shading marks the region meeting the hERG and AMES limits. (b) Four predictions divided by their task upper limits. Connected pairs compare B2-09 and B3-02 at each endpoint; ratios at or below one meet that target.

Panel of Figure 7 — Figure 7: Scaffold replacement and structural inspection. (a) The reference ligand and submitted candidate have different ring scaffolds and share a sulfonamide group. (b) The predicted Boltz-2 complex was retrieved and used as the code’s sole CIF input. For each selected residue $r$ , the script computed the minimum distance to an atom of the predicted ligand $L$ . The full 21-residue report was read before the same candidate was submitted. Arrows follow the recorded scientific handoffs.

Panel of Figure 8 — Figure 8: Local static geometry of the predicted complex. Panel (b) shows the nearest-ligand distances recorded during task execution. The pocket overlay in (a) uses structures aligned after execution for visualization. Orange points identify His94 and Thr198; the remaining selected residues are blue.

Panel of Figure 9 — Figure 9: QED feedback reused during molecular optimization. (a) The source, first- and second-batch leaders, and submitted candidate, drawn from the recorded molecular structures. Arrows indicate reuse of evaluated structures as baselines in subsequent scripts. (b) All 80 numerical entries from batches containing 26, 26, and 31 records, including repeated structures. Orange, light blue, and dark blue indicate batches 1–3. Diamonds mark batch maxima, the star marks the submitted candidate, and the dashed line denotes the source QED. Three invalid first-batch entries occupy a separate non-numerical strip. The submitted candidate shares the highest third-batch QED.

Source task	Records scanned	$K_{i}$ (A), nM	$K_{i}$ (B), nM
003	2,008	7.9	2,512
021	195	9.5	2,300

Appendix

Table 8: Affinity records recovered by the first independent review. The reviewer matched both candidate structures in each source task; A/B labels are task-specific.

Figure 10: From scientific feedback to revised affinity guidance. Independent review results and researcher-selected execution feedback informed the S0-to-S1 revision. The added recovery-and-matching procedure was exercised in a second review on 003 and 021 before its use on task 012.

Figure 11: CXCR2 candidates and matched affinity records. Molecular structures are task inputs;

K_{i}

values were retrieved from BindingDB records associated with Karlström et al. (2013) .

Comparison	Initial skill (S0)	Revised skill (S1)
Source processing	Partial source reads	Full activity file recovered; 1,523 records scanned
Decision evidence	Property or conformer heuristics	One matched record per molecule; corresponding $K_{i}$ values
Selected molecule	B (incorrect)	A (correct)

Appendix

Table 9: Evidence use and candidate selection in task 012. The initial skill (S0) and revised skill (S1) were each used in one execution of the CXCR2 affinity task, which asks for the weaker binder (higher

K_{i}

LLM Agent Harnesses

LLM agent harnesses are the software and runtime systems that organize an agent’s tools, context, state, execution, and safeguards.

See all 377 papers

MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization
Apr 2, 2026Lisheng Zhang, Lilong Wang, Xiangyu Sun +14Drug Discovery Autonomous Research Agents
Peking University Health Science Center, Peking University, Beijing, China · Shanghai AI Laboratory, Shanghai, China · Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China
HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses
Sep 14, 2026Jieyuan Liu, Mengzhou Hu, Jefferson Chen +10Multi-Agent LLM Systems Scientific Hypothesis Generation
University of California San Diego · 2Texas A&M University · 3Carnegie Mellon University +1
MADDIE: Multi-Agent Drug Discovery Engine
Oct 12, 2026Ye Liu, Reza Averly, Avery Meyer +4Drug Discovery Multi-Agent LLM Systems
Biomedical Informatics, The Ohio State University. · Computer Science and Engineering, The Ohio State University. · Medicinal Chemistry and Pharmacognosy, The Ohio State University. +1

Apr 2, 2026J/K move · Enter open · S save

MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization

Lisheng Zhang, Lilong Wang, Xiangyu Sun, Wei Tang+13

Peking University Health Science Center, Peking University, Beijing, China · Shanghai AI Laboratory, Shanghai, China · Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China

Drug Discovery Autonomous Research Agents Molecular Optimization LLM Agent Orchestration AI Agent Benchmarks Agent Skill Learning

Open paper AskPDF arXiv

Computational drug discovery, particularly the complex workflows of drug molecule screening and optimization, requires orchestrating dozens of specialized tools in multi-step workflows, yet current AI agents struggle to maintain robust performance and consistently underperform in these high-complexity scenarios. Here we present MolClaw, an autonomous agent that leads drug molecule evaluation, screening, and optimization. It unifies over 30 specialized domain resources through a three-tier hierarchical skill architecture (70 skills in total) that facilitates agent long-term interaction at runtime: tool-level skills standardize atomic operations, workflow-level skills compose them into validated pipelines with quality check and reflection, and a discipline-level skill supplies scientific principles governing planning and verification across all scenarios in the field. Additionally, we introduce MolBench, a benchmark comprising molecular screening, optimization, and end-to-end discovery challenges spanning 8 to 50+ sequential tool calls. MolClaw achieves state-of-the-art performance across all metrics, and ablation studies confirm that gains concentrate on tasks that demand structured workflows while vanishing on those solvable with ad hoc scripting, establishing workflow orchestration competence as the primary capability bottleneck for AI-driven drug discovery.

Abstract

LLM Agent Harnesses

Explore similar work

MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization

HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses

MADDIE: Multi-Agent Drug Discovery Engine