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

TL;DR

M^4olGen tackles the challenge of exact multi-property molecular generation by combining retrieval-augmented prototyping with fragment-level optimization guided by GRPO. The two-stage design first constructs a chemically valid prototype near the target region and then performs controlled, multi-hop edits at the fragment level to minimize distance to multi-property targets such as QED, LogP, MW, and HOMO/LUMO. The work contributes a scalable BRICS-based dataset with neighbor edit relations to train the GRPO policy, demonstrates substantial improvements over strong LLMs and graph baselines, and achieves precise target satisfaction while preserving validity, diversity, and uniqueness. This approach offers a practical pathway to scalable, numerically controlled molecular design with potential impact on drug and material discovery, and provides a public resource to support fragment-level learning and reasoning in chemical design.

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.

Figures (3)

• Figure 1: The flow chart of M$^{4}$olGen. The first two blocks involve Retrieval and Prototyping, where molecular candidates are first retrieved based on the given constraints (QED, LogP, MW) and then analyzed by a local reasoner to extract constraints, analyze retrieved molecules, and propose an editing plan based on evaluator's feedback to generate prototypes iteratively. The third block describes Multi-Hop Optimization, where the prototypes are optimized through one-hop and n-hop controllable editing steps by the molecule optimizer trained by GRPO.
• Figure 2: Ablation curves showing the drop percentage (higher is better) of each error metric relative to the no-retrieval baseline across methods. Curves are shown for QED, logP, MW, and the normalized total error.
• Figure 3: The demo of nodes and edges of molecule neighbor reasoning dataset.