Multi-perspective explainer reveals how graph models predict molecules
MPGE: A Multi-Perspective Graph Explainer for Molecular Classification Explanation
Machine LearningArtificial Intelligence
Summary
Graph neural networks can predict the properties of molecules using their chemical structures, but it's hard to understand why they make a certain decision. The authors propose a new explanation method called MPGE that shows what parts of the molecule support the prediction, what small changes could change the result, and what variations the model accepts without changing its mind. This approach helps clarify the model’s reasoning by providing three different views of the explanation. The method was tested on several molecular datasets and was able to highlight important structural features with fewer edges.
What this means in practice
- •For drug discovery teams: Use MPGE to understand which molecular features influence AI-predicted drug properties, improving trust in molecular screening models.
- •For chemical informatics developers: Incorporate MPGE into cheminformatics tools to generate interpretable molecular explanations supporting AI-driven property predictions.
Authors
Mahtab Sarvmaili
Abstract
Graph neural networks (GNNs) predict molecular properties from chemical graph data, but predictive accuracy does not explain how graph information supports an individual decision. A compact prediction-preserving rationale does not necessarily reveal which changes reverse the decision or which modifications the model tolerates. We propose the Multi-Perspective Graph Explainer (MPGE), unifying factual support, counterfactual sensitivity, and exemplar tolerance for a frozen classifier. The factual view, originally termed prototype (PT), seeks a compact retained edge set with the same label and required confidence. Counterfactual (CF) explanations seek bounded prediction-changing deletions; exemplar (EXE) explanations seek non-trivial bounded deletions that preserve the label and confidence. A shared constrained formulation connects prediction behavior, compactness, and edit cost, while separate objectives generate the three views. Our graph-classification extension of CF-GNNExplainer learns symmetric edge rankings and verifies discrete candidates, recording unsuccessful searches. A separate BBBP fragment backend returns RDKit-sanitized molecules. We evaluate the primary GCN implementation on MUTAG, Mutagenicity, AIDS, COX2_MD, and BBBP using semantic coverage, conditional quality, stability, and runtime. Successful factual masks retained 8.6%--15.5% of input edges on average across datasets; bounded counterfactual coverage was 4.8%--67.6%, and exemplar preservation coverage was 98.9%--100.0%. Exploratory controls reveal the influence of hard projection and retained node information. Quantitative comparisons and molecular visualizations characterize model support, sensitivity, and tolerance without treating them as validated chemical mechanisms.