Fragment graphs improve molecular predictions and interpretation for drug discovery
Fragment-Level Graph Encoding for Interpretable Molecular Property Prediction
Machine Learning
Summary
It is hard for chemists to understand why certain AI models predict properties of molecules when these models focus on single atoms. The authors suggest representing molecules as smaller pieces called fragments instead of individual atoms, so explanations are clearer without losing prediction accuracy. This method works with existing graph neural networks and helps highlight important groups of atoms related to drug activity more reliably. Their approach makes it easier for chemists to connect AI insights with known chemical knowledge.
What this means in practice
- •For medicinal chemists: Use fragment-level molecular graphs to get clearer functional group explanations from AI predictions aiding drug design decisions.
- •For chemical software developers: Develop molecular property prediction tools that apply fragment-based graphs for improved interpretability without losing accuracy.
Authors
Rafael Lopes Almeida, Ina Pöhner
Abstract
Graph neural networks (GNNs) are among the standard tools for molecular property prediction, but the atom-level explanations of their predictions are hard for medicinal chemists to act on. We propose a change of input representation rather than a new model: molecules are encoded as fragment-level graphs from BRICS decomposition, where each node is a synthetically meaningful fragment whose features are the element-wise sum of its atoms' features. Because only the representation changes, any message-passing GNN and any post-hoc attribution method produces explanations at the resolution of functional groups instead of single atoms. Across four architectures, four prediction tasks, and two explainers under scaffold splits, fragment-level encoding matches atom-level test accuracy on roughly five-fold smaller graphs. Integrated-Gradients attributions recover known pharmacophores above chance on two enzyme targets, namely BACE at 1.5 times the chance rate, and PTR1 at 3.2 times. A like-for-like comparison against an aggregated atom-level baseline shows that post-hoc aggregation is not a stable substitute: its BACE hit rate range both exceeds and falls below the fragment model's hit rate depending on two arbitrary aggregation choices, whereas fragment-level training yields a single convention-free attribution at no predictive cost. The code is available at: https://github.com/4D-Lab/frame