MFG: Multi-turn Feedback Guided Molecular Optimization
Abstract
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. To address this gap, we introduce Multi-turn Feedback Guided (MFG) molecular optimization, a reinforcement learning framework that trains an LLM to progressively optimize molecules using feedback from previous turns. MFG introduces a feedback vector capturing both property improvements and source molecule similarity, alongside a feedback-aligned reward function that promotes progressive improvement across turns. We apply MFG over GeLLM3O, a state-of-the-art LLM model for multi-objective molecular optimization, evaluate GeLLM3O−MFG on ten multi-objective molecular optimization tasks. The results demonstrate that GeLLM3O−MFG achieves the highest average similarity-constrained success rate (81.50%), outperforming the base model GeLLM3O (70.81%).
Related
- Reinforcement learning with evolutionary priors unlocks multi-site enzyme optimization from sparse experimental feedback
- OmicsBench: a long-horizon multi-turn benchmark of AI agents on research-study-scale bioinformatics
- Enhancing two-dimensional control via single-channel haptic feedback using a multi-dimensional encoding strategy
- Synergistic Effects of Behavioral Feedback and Seasonality Generate Chaos in Cooperative Multi-Pathogen Systems
- Active-learning-guided optimization of cell-free systems for genome-wide transcriptomic profiling reveals progressive layers of regulation
BibTeX
@misc{nguyen2026multiturn,
title = {{MFG: Multi-turn Feedback Guided Molecular Optimization}},
author = {Trieu Nguyen and Frazier N. Baker and Ruoxi Gao and Daniel Adu-Ampratwum and Xia Ning},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.18851},
eprint = {2610.18851},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.18851},
note = {CC BY-SA 4.0},
}
RIS
TY - UNPB
TI - MFG: Multi-turn Feedback Guided Molecular Optimization
AU - Trieu Nguyen
AU - Frazier N. Baker
AU - Ruoxi Gao
AU - Daniel Adu-Ampratwum
AU - Xia Ning
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.18851
UR - https://arxiv.org/abs/2610.18851
AB - 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. To address this gap, we introduce Multi-turn Feedback Guided (MFG) molecular optimization, a reinforcement learning framework that trains an LLM to progressively optimize molecules using feedback from previous turns. MFG introduces a feedback vector capturing both property improvements and source molecule similarity, alongside a feedback-aligned reward function that promotes progressive improvement across turns. We apply MFG over GeLLM^3O, a state-of-the-art LLM model for multi-objective molecular optimization, evaluate GeLLM^3O−MFG on ten multi-objective molecular optimization tasks. The results demonstrate that GeLLM^3O−MFG achieves the highest average similarity-constrained success rate (81.50%), outperforming the base model GeLLM^3O (70.81%).
ER -