ETALON: Verified Execution and Persistent Research Memory for Agentic Computer-Aided Drug Design
Language-model agents can coordinate computational drug discovery, but adapting a campaign to new evidence requires scientific software that supports flexible workflows, consistent modeling, and continuity across research sessions. Here we present ETALON, a harness that equips a general-purpose coding agent with an integrated environment for computer-aided drug design. Three independently usable systems supply its scientific operations: MolQuarry for chemical and biological evidence, MolCascade for molecular preparation, screening, and docking, and PRISM for molecular generation and physical modeling. Each exposes configurable procedures with explicit scientific requirements, so the host can revise its research strategy while retaining consistent preparation and evaluation practices. Operations registered with a common execution layer are checked before dispatch, and their outputs carry separate records of completion, verification, and admission, so that an incomplete computation is not recorded as an unfavorable molecular result. A persistent research graph links evidence, judgments, experiments, and results across sessions, and candidate method revisions are compared with frozen baselines under criteria declared in advance.
Authors
- Zaixing Yang (ORCID: https://orcid.org/0000-0003-3521-6867)
- Xufan Gao (ORCID: https://orcid.org/0009-0004-0065-3172)
- Mao Xiyuan
- Xuanyi Zhu
- Yifei Chen
- Zhaoqi Shi
- Mingyu Xu
- Xin Zheng
Institutions
- Institute of Biophysics (CN)
- Zhejiang University (CN)
Publication Details
- Journal
- ChemRxiv
- Published
- 2026-11-02
- DOI
- https://doi.org/10.26434/chemrxiv.15011490/v1
- Primary Topic
- Computational Drug Discovery Methods
- Type
- preprint