MolReasoner: Toward Effective and Interpretable Reasoning for Molecular LLMs

2025-08-04

AI Summary

Existing large language models (LLMs) suffer from domain-specific semantic gaps, poor interpretability, and insufficient reasoning depth in molecular reasoning. To address these challenges, we propose MolReasoner—a two-stage framework. In the first stage, we perform supervised fine-tuning (Mol-SFT) using high-quality, GPT-4o-generated and human-verified molecular chain-of-thought (CoT) data. In the second stage, we introduce Mol-RL: a reinforcement learning phase guided by a multi-dimensional reward function that jointly incorporates chemical rules and model-based judgments to achieve deep structural–linguistic alignment. This work is the first to integrate high-fidelity synthetic CoT with a chemistry-aware RL reward mechanism, significantly enhancing both interpretability and generalization. MolReasoner consistently outperforms baselines across molecular property prediction, reaction prediction, and other downstream tasks, advancing LLMs from memorization-driven generation toward principled, interpretable chemical reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Language Models(LLMs) have demonstrated remarkable performance across various domains, yet their capabilities in molecular reasoning remain insufficiently explored. Current approaches tend to rely heavily on general-purpose prompting, which lacks domain-specific molecular semantics, while those that use fine-tuning strategies often face challenges with interpretability and reasoning depth. To address these issues, we introduce MolReasoner, a two-stage framework designed to transition LLMs from memorization towards chemical reasoning. First, we propose Mol-SFT, which initializes the model's reasoning abilities via synthetic Chain-of-Thought(CoT) samples generated by GPT-4o and verified for chemical accuracy. Subsequently, Mol-RL applies reinforcement learning with specialized reward functions designed explicitly to align chemical structures with linguistic descriptions, thereby enhancing molecular reasoning capabilities. Our approach notably enhances interpretability, improving the model 's molecular understanding and enabling better generalization. Extensive experiments demonstrate that MolReasoner outperforms existing methods, and marking a significant shift from memorization-based outputs to robust chemical reasoning.

Problem

• Enhancing molecular reasoning in LLMs with domain-specific semantics
• Improving interpretability and depth in molecular fine-tuning strategies
• Transitioning from memorization to robust chemical reasoning in LLMs

Innovation

• Two-stage framework for molecular reasoning enhancement
• Synthetic CoT samples for model initialization
• Reinforcement learning with specialized reward functions