M^4olGen: Multi-Agent, Multi-Stage Molecular Generation under Precise Multi-Property Constraints

2026-01-15

AI Summary

This work addresses the challenge of generating valid molecules under precise multi-objective physicochemical constraints in drug design. The authors propose a two-stage, fragment-based molecular generation framework: in the first stage, a multi-agent retrieval-augmented reasoning process generates prototype molecules proximal to the feasible region; in the second stage, Group Relative Policy Optimization (GRPO) performs fine-grained molecular editing to accurately approach target properties while controlling structural deviation. By integrating multi-agent reasoning with fragment-level reinforcement learning— a novel combination— the method significantly outperforms existing large language models and graph-based generative approaches on tasks involving simultaneous constraints on QED, LogP, molecular weight, and HOMO/LUMO energy levels. The framework achieves controllable, interpretable, and reproducible high-precision molecular generation.

Technology Category

• Search and Optimization: Learning to Search
• Multiagent Systems: Mechanism Design
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Generating molecules that satisfy precise numeric constraints over multiple physicochemical properties is critical and challenging. Although large language models (LLMs) are expressive, they struggle with precise multi-objective control and numeric reasoning without external structure and feedback. We introduce \textbf{M olGen}, a fragment-level, retrieval-augmented, two-stage framework for molecule generation under multi-property constraints. Stage I : Prototype generation: a multi-agent reasoner performs retrieval-anchored, fragment-level edits to produce a candidate near the feasible region. Stage II : RL-based fine-grained optimization: a fragment-level optimizer trained with Group Relative Policy Optimization (GRPO) applies one- or multi-hop refinements to explicitly minimize the property errors toward our target while regulating edit complexity and deviation from the prototype. A large, automatically curated dataset with reasoning chains of fragment edits and measured property deltas underpins both stages, enabling deterministic, reproducible supervision and controllable multi-hop reasoning. Unlike prior work, our framework better reasons about molecules by leveraging fragments and supports controllable refinement toward numeric targets. Experiments on generation under two sets of property constraints (QED, LogP, Molecular Weight and HOMO, LUMO) show consistent gains in validity and precise satisfaction of multi-property targets, outperforming strong LLMs and graph-based algorithms.

Problem

• molecular generation
• multi-property constraints
• precise numeric control
• physicochemical properties

Innovation

• fragment-level editing
• multi-agent reasoning
• retrieval-augmented generation
• Group Relative Policy Optimization (GRPO)
• multi-property constrained optimization