Large Language Models for Controllable Multi-property Multi-objective Molecule Optimization
Vishal Dey, Xiaohua Hu, Xia Ning · 2025
In real-world drug design, molecule optimization requires selectively improving multiple molecular properties up to pharmaceutically relevant levels, while maintaining others that already meet such criteria.However, existing computational approaches and instructiontuned LLMs fail to capture such nuanced property-specific objectives, limiting their practical applicability.To address this, we introduce C-MuMOInstruct, the first instructiontuning dataset focused on multi-property optimization with explicit, property-specific objectives.Leveraging C-MuMOInstruct, we develop GeLLM 4 O-Cs, a series of instructiontuned LLMs that can perform targeted propertyspecific optimization.Our experiments across 5 in-distribution and 5 out-of-distribution tasks show that GeLLM 4 O-Cs consistently outperform strong baselines, achieving up to 126% higher success rate.Notably, GeLLM 4 O-Cs exhibit impressive 0-shot generalization to novel optimization tasks and unseen instructions.This offers a step toward a foundational LLM to support realistic, diverse optimizations with property-specific objectives.C-MuMOInstruct and code are accessible through https:// github.com/ninglab/GeLLMO-C.