发表机构
Dalhousie University（达尔豪斯大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究提出多视角图解释器MPGE，针对冻结分类器统一事实支撑、反事实敏感性和exemplar容忍度，在多个分子数据集上验证了其解释性能。
AI 中文摘要
图神经网络（GNN）可从化学图数据预测分子性质，但预测精度无法解释图信息如何支撑单个决策。紧凑的预测保留型理由未必能揭示哪些变更会反转决策，或模型可容忍哪些修改。我们提出多视角图解释器（MPGE），针对冻结分类器统一了事实支撑、反事实敏感性和 exemplar 容忍度。事实视角（原称原型（PT））寻求具有相同标签和所需置信度的紧凑保留边集；反事实（CF）解释寻求有界的预测变更删除；exemplar（EXE）解释寻求保留标签和置信度的非平凡有界删除。共享的约束公式连接了预测行为、紧凑性和编辑成本，而独立的目标生成三个视角。我们对 CF-GNNExplainer 进行图分类扩展，学习对称边排名并验证离散候选，记录未成功的搜索。单独的 BBBP 片段后端返回经 RDKit 净化的分子。我们在 MUTAG、致突变性（Mutagenicity）、AIDS、COX2_MD 和 BBBP 上评估主要的 GCN 实现，使用语义覆盖率、条件质量、稳定性和运行时间指标。成功的事实掩码在各数据集上平均保留了输入边的 8.6%--15.5%；有界反事实覆盖率为 4.8%--67.6%，exemplar 保留覆盖率为 98.9%--100.0%。探索性控制揭示了硬投影和保留节点信息的影响。定量比较和分子可视化表征了模型的支撑、敏感性和容忍度，且未将其视为已验证的化学机制。
英文摘要
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.
