SMILESGNN：SMILES-图交叉注意力融合的可解释临床毒性预测
原标题：SMILESGNN: Interpretable Clinical Toxicity Prediction via SMILES-Graph Cross-Attention Fusion
AI 摘要
研究者提出 SMILESGNN，一种通过交叉注意力融合 SMILES Transformer 编码器与 GATv2 图编码器的多模态架构，用于药物毒性预测，并保留显式图分支以支持 GNNExplainer 解释。在 ClinTox 上以仅 0.4M 参数取得 AUC-ROC 0.987、F1 0.906，优于所有图单模态基线；其预训练变体 SMILESGNN-PT 在 Tox21 的 12 项任务上平均 AUC-ROC 为 0.750。结果表明交叉注意力是兼顾预测性能与图可解释性的实用融合方案。
正文节选
SMILESGNN: Interpretable Clinical Toxicity Prediction via SMILES-Graph Cross-Attention Fusion Abstract Drug toxicity prediction is critical for reducing late-stage attrition in drug discovery, yet remains challenging due to severe class imbalance, scaffold-based generalization, and the clinical need for interpretable predictions. Single-modality approaches-SMILES Transformers or graph neural networks-capture complementary aspects of molecular structure, while sequence-only models cannot directly