Reasoning-Enhanced Large Language Models for Molecular Property Prediction

TL;DR

MPPReasoner introduces a chemical reasoning capability into a multimodal large language model for molecular property prediction by integrating 2D molecular images and SMILES strings. The authors employ a two-stage training regime—supervised fine-tuning with 16,000 expert-curated reasoning trajectories followed by reinforcement learning from principle-guided rewards (RLPGR)—to cultivate domain-specific reasoning. Across 8 datasets, MPPReasoner achieves notable improvements and exceptional cross-task generalization, with strong ID and OOD performance and chemically sound, expert-validated reasoning paths. This work advances interpretable AI in chemistry and offers a blueprint for embedding structured domain reasoning into domain-specific tasks within LLM frameworks.

Abstract

Molecular property prediction is crucial for drug discovery and materials science, yet existing approaches suffer from limited interpretability, poor cross-task generalization, and lack of chemical reasoning capabilities. Traditional machine learning models struggle with task transferability, while specialized molecular language models provide little insight into their decision-making processes. To address these limitations, we propose \textbf{MPPReasoner}, a multimodal large language model that incorporates chemical reasoning for molecular property prediction. Our approach, built upon Qwen2.5-VL-7B-Instruct, integrates molecular images with SMILES strings to enable comprehensive molecular understanding. We develop a two-stage training strategy: supervised fine-tuning (SFT) using 16,000 high-quality reasoning trajectories generated through expert knowledge and multiple teacher models, followed by Reinforcement Learning from Principle-Guided Rewards (RLPGR). RLPGR employs verifiable, rule-based rewards that systematically evaluate chemical principle application, molecular structure analysis, and logical consistency through computational verification. Extensive experiments across 8 datasets demonstrate significant performance improvements, with MPPReasoner outperforming the best baselines by 7.91\% and 4.53\% on in-distribution and out-of-distribution tasks respectively. MPPReasoner exhibits exceptional cross-task generalization and generates chemically sound reasoning paths that provide valuable insights into molecular property analysis, substantially enhancing both interpretability and practical utility for chemists. Code is available at https://anonymous.4open.science/r/MPPReasoner-12687.

Figures (6)

• Figure 1: Overview of MPPReasoner framework.
• Figure 2: Illustration of RLPGR in MPPReasoner.
• Figure 3: Reasoning quality evaluation results. (a) Strong consistency between automated and human assessments with $\rho$ = 0.82. (b) MPPReasoner achieves the highest reasoning quality score.
• Figure 4: Case study comparison between GPT-4o and MPPReasoner for CY-P450-2C9 substrate prediction. The figure demonstrates how RLPGR training enables systematic chemical reasoning with accurate predictions, while GPT4o suffer from several errors.
• Figure 5: Successful case on MPPReasoner for BACE1 protein binding prediction (ID).