Empowering Machines to Think Like Chemists: Unveiling Molecular Structure-Polarity Relationships with Hierarchical Symbolic Regression

TL;DR

This paper introduces UHiSR, a novel method combining neural networks and symbolic regression to automatically derive interpretable equations linking molecular structure to polarity, enhancing understanding in chromatography analysis.

Contribution

The paper presents UHiSR, a new approach that automatically generates interpretable models connecting molecular structure with polarity, addressing the interpretability-expressiveness trade-off in AI models for chromatography.

Findings

UHiSR successfully distills chemical-intuitive polarity indices.

It discovers interpretable equations linking molecular structure to chromatographic behavior.

The method improves interpretability of molecular polarity models.

Abstract

Thin-layer chromatography (TLC) is a crucial technique in molecular polarity analysis. Despite its importance, the interpretability of predictive models for TLC, especially those driven by artificial intelligence, remains a challenge. Current approaches, utilizing either high-dimensional molecular fingerprints or domain-knowledge-driven feature engineering, often face a dilemma between expressiveness and interpretability. To bridge this gap, we introduce Unsupervised Hierarchical Symbolic Regression (UHiSR), combining hierarchical neural networks and symbolic regression. UHiSR automatically distills chemical-intuitive polarity indices, and discovers interpretable equations that link molecular structure to chromatographic behavior.