MPGE: A Multi-Perspective Graph Explainer for Molecular Classification Explanation
Graph neural networks (GNNs) predict molecular properties from chemical graph data, but predictive accuracy does not explain how graph information supports an individual decision.
ProofPaper ↗
Key points
- We propose the Multi-Perspective Graph Explainer (MPGE), unifying factual support, counterfactual sensitivity, and exemplar tolerance for a frozen classifier.
- Our graph-classification extension of CF-GNNExplainer learns symmetric edge rankings and verifies discrete candidates, recording unsuccessful searches.
- We evaluate the primary GCN implementation on MUTAG, Mutagenicity, AIDS, COX2MD, and BBBP using semantic coverage, conditional quality, stability, and runtime.
- Quantitative comparisons and molecular visualizations characterize model support, sensitivity, and tolerance without treating them as validated chemical mechanisms.
Sources (1)
- [1]MPGE: A Multi-Perspective Graph Explainer for Molecular Classification ExplanationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 02:28 PM
Graph neural networks (GNNs) predict molecular properties from chemical graph data, but predictive accuracy does not explain how graph information supports an individual decision.
We propose the Multi-Perspective Graph Explainer (MPGE), unifying factual support, counterfactual sensitivity, and exemplar tolerance for a frozen classifier.
Extractive summary: sentences quoted from the sources.