DISSOLVR: An Interpretable and Fast Framework for Aqueous and Organic Solubility Prediction

2026-10-01

AI Summary

This study addresses the lack of physical interpretability in deep learning-based solubility prediction by proposing a transparent predictive framework. Methodologically, it employs physically grounded molecular descriptors as the basis for feature mapping and integrates large language model (LLM)-driven post hoc explanation techniques to generate chemical narratives, thereby effectively overcoming the traditional black-box paradigm. Results demonstrate that this framework achieves predictive accuracy approaching experimental error limits while exhibiting superior out-of-distribution generalization capabilities, earning strong endorsement from domain experts. By reconciling high predictive accuracy with robust transparency, this work establishes a novel paradigm for applying interpretable artificial intelligence in chemistry.

Abstract

High-fidelity solubility prediction is fundamental to pharmaceutical development and environmental partitioning, where accurate modeling must couple molecular structure with thermodynamic behavior across diverse chemical environments. However, recent advancements have been dominated by deep learning architectures that often sacrifice physical interpretability for predictive power. We challenge this trend by showing that state-of-the-art performance does not require such non-transparent architectures. To address this, we introduce DISSOLVR, a transparent framework for molecular solubility prediction. In addition, we perform a comprehensive literature review and a benchmarking study against various methods. We show that DISSOLVR approaches the aleatoric limit of experimental uncertainty and achieves OOD generalization through structural invariance, derived by mapping molecules to physically-grounded descriptors. Then, we present an LLM-assisted post-hoc explanation pipeline that bridges the gap between symbolic model artifacts and chemically grounded narratives. Finally, a comparative benchmark of a survey involving 22 expert chemists reveals that expert evaluators provide deep insights.

Problem

• solubility prediction
• physical interpretability
• deep learning
• molecular structure
• out-of-distribution generalization

Innovation

• Solubility Prediction
• Interpretable Framework
• Out-of-Distribution Generalization
• Physically-grounded Descriptors
• LLM-assisted Explanation