M3OS: A Monte Carlo Graph Search-Orchestrated Multi-Agent LLM System for Evidence-Traced Molecular Optimization
Abstract
Small-molecule optimization integrates medicinal-chemistry reasoning and computational evidence through iterative, multi-objective decisions. When large language models (LLMs) reason over optimization histories stored primarily in conversational context, they must recover candidate identities, prior evaluations, and task constraints to guide subsequent decisions. We present M3OS, a multi-agent LLM system that decouples molecular-design reasoning from optimization-state management through Monte Carlo graph search. A persistent graph links evaluated candidates, parent-child transformations and evaluation evidence, while rewards and visit statistics guide LLM-assisted parent selection. Two branches combine tool-driven candidate generation with knowledge- and case-guided medicinal-chemistry editing. An execution harness controls graph updates through structured output extraction, molecular validation and task-bound evaluation. Agents receive role-specific contexts, while the graph preserves optimization trajectories beyond their active contexts. Across three molecular optimization benchmarks, M3OS achieves higher success rates than baselines, supporting the integration of persistent search state, specialized agents and controlled execution for multi-constraint optimization.
Related
- Multivalent Surface Search Dynamics Shape Bacteriophage Adsorption Efficiency: A Stochastic Model of Tail Fiber Optimization
- Editorial: Green biomanufacturing by industrial microorganisms: precise regulation and system optimization
- 1017 Computational design of de novo antigen-specific T cell receptors using AI optimization and long-time-scale molecular simulations for targeted cancer immunotherapy
- Expanding the clinical and molecular spectrum of Schaaf-Yang syndrome in China: evidence for a severe phenotypic subgroup associated with MAGEL2 c.1996-Region variants
- Extraction, Structure–Activity Relationships, and Modification of Plant-Derived Polysaccharides: An Integrated Framework for Evidence-Based Method Selection and Bioactivity Optimization—A Review
BibTeX
@misc{wang2026m3os,
title = {{M3OS: A Monte Carlo Graph Search-Orchestrated Multi-Agent LLM System for Evidence-Traced Molecular Optimization}},
author = {Junjie Wang and Yaowei Jin and Ruohui Tang and Guonan Cui and Haojie Wang and Penglei Wang and Dingyan Wang and Duo An and Shuangjia Zheng and Qian Shi},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.34491},
eprint = {2609.34491},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.34491},
}
RIS
TY - UNPB
TI - M3OS: A Monte Carlo Graph Search-Orchestrated Multi-Agent LLM System for Evidence-Traced Molecular Optimization
AU - Junjie Wang
AU - Yaowei Jin
AU - Ruohui Tang
AU - Guonan Cui
AU - Haojie Wang
AU - Penglei Wang
AU - Dingyan Wang
AU - Duo An
AU - Shuangjia Zheng
AU - Qian Shi
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.34491
UR - https://arxiv.org/abs/2609.34491
AB - Small-molecule optimization integrates medicinal-chemistry reasoning and computational evidence through iterative, multi-objective decisions. When large language models (LLMs) reason over optimization histories stored primarily in conversational context, they must recover candidate identities, prior evaluations, and task constraints to guide subsequent decisions. We present M3OS, a multi-agent LLM system that decouples molecular-design reasoning from optimization-state management through Monte Carlo graph search. A persistent graph links evaluated candidates, parent-child transformations and evaluation evidence, while rewards and visit statistics guide LLM-assisted parent selection. Two branches combine tool-driven candidate generation with knowledge- and case-guided medicinal-chemistry editing. An execution harness controls graph updates through structured output extraction, molecular validation and task-bound evaluation. Agents receive role-specific contexts, while the graph preserves optimization trajectories beyond their active contexts. Across three molecular optimization benchmarks, M3OS achieves higher success rates than baselines, supporting the integration of persistent search state, specialized agents and controlled execution for multi-constraint optimization.
ER -