Summary
MFG (Multi-turn Feedback Guided) molecular optimization is a reinforcement learning framework that trains a large language model to progressively optimize molecules across multiple rounds using feedback from previous turns. Multi-objective molecular optimization, a key step in drug discovery, seeks to improve several molecular properties simultaneously while keeping the overall structure similar, typically through iterative modification guided by computational or experimental feedback. Existing LLM-based methods were not explicitly trained to exploit such feedback and mostly target single-turn optimization. MFG addresses this gap by introducing feedback vectors that capture both property improvement and similarity to the source molecule, combined with a feedback-aligned reward function that encourages progressive improvement across turns. Applied to GeLLM3O, a state-of-the-art LLM for multi-objective molecular optimization, the resulting GeLLM3O-MFG model was evaluated on ten multi-objective optimization tasks and achieved the highest average similarity-constrained success rate of 81.50%, surpassing the base GeLLM3O model's 70.81%. Paper: arXiv 2610.18851.
Paper Overview
Field: Machine Learning
Authors: Trieu Nguyen, Frazier N. Baker, Ruoxi Gao, Daniel Adu-Ampratwum, Xia Ning
Published: 2026-10-13
arXiv: 2610.18851
Summary
Multi-objective molecular optimization is a critical process in drug discovery that aims to modify a given molecule to simultaneously improve multiple properties while maintaining a similar overall structure. This process often requires iterative modifications guided by computational or experimental feedback. Recent large language model (LLM)-based methods have begun to explore this iterative approach; however, these methods are not explicitly trained to leverage such feedback, while existing approaches primarily focus on single-turn molecular optimization.
Key Contributions
- MFG framework: a reinforcement learning framework that trains an LLM to progressively optimize molecules using feedback from previous turns.
- Feedback vectors: capture both property improvement and similarity to the source molecule.
- Feedback-aligned reward: a reward function promoting progressive improvement across turns.
Results
Applied to GeLLM3O, a state-of-the-art multi-objective molecular optimization LLM, the resulting GeLLM3O-MFG was evaluated on ten multi-objective tasks and achieved the highest average similarity-constrained success rate of 81.50%, surpassing the base GeLLM3O model (70.81%).
Links
- Paper: https://arxiv.org/abs/2610.18851
---
*Auto-collected on 2026-10-15*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178635995