MFG: Multi-turn Feedback Guided Molecular Optimization
arXiv cs.AI · 14 Oct 2026, 03:58 AEST
Multi-turn Feedback Guided (MFG) molecular optimization enhances drug discovery by iteratively modifying molecules based on feedback. This new framework utilizes a large language model to improve multiple properties while maintaining structural similarity.
Key facts
- 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.
- 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 $ ext{GeLLM}^{3} ext{O}$, a state-of-the-art LLM model for multi-objective molecular optimization, evaluate $ ext{GeLLM}^{3} ext{O-MFG}$ on ten multi-objective molecular optimization tasks.
- The results demonstrate that $ ext{GeLLM}^{3} ext{O-MFG}$ achieves the highest average similarity-constrained success rate ($81.50%$), outperforming the base model $ ext{GeLLM}^{3} ext{O}$ ($70.81%$).
Read aloud
Multi-turn Feedback Guided MFG molecular optimization enhances drug discovery by iteratively modifying molecules based on feedback. This new framework utilizes a large language model to improve multiple properties while maintaining structural similarity. 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. To address this gap, we introduce Multi-turn Feedback Guided MFG molecular optimization, a reinforcement learning framework that trains an L L M 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 $ ext GeLLM ^ three ext O $, a state-of-the-art L L M model for multi-objective molecular optimization, evaluate $ ext GeLLM ^ three ext O-MFG $ on ten multi-objective molecular optimization tasks. The results demonstrate that $ ext GeLLM ^ three ext O-MFG $ achieves the highest average similarity-constrained success rate eighty one dollars and fifty cents%$, outperforming the base model $ ext GeLLM ^ three ext O $ seventy dollars and eighty one cents%$.
Sources in this cluster
- MFG: Multi-turn Feedback Guided Molecular Optimization
arXiv cs.AI · 14 Oct 2026, 03:58 AEST