Scholarlink

Accurate drug-drug interaction (DDI) prediction is crucial, yet current graph neural network (GNN) methods face limitations. Their fixed propagation and aggregation schemes struggle to differentiate structurally dissimilar multisets sharing elements, impairing feature extraction and capturing complex graph topologies. Additionally, relying solely on 2D chemical structures fails to reveal complex biological activities or link molecules with different structures but similar effects, limiting prediction accuracy. To overcome these challenges, we propose an improved GCN model incorporating 3D molecular similarity. Our approach innovatively uses a message-passing and aggregation scheme based on neighborhood overlap similarity and Pearson correlation coefficient. We also integrate 3D molecular similarity features during feature fusion. This enhanced model better captures complex drug interaction relationships, achieving a 3% to 16% accuracy improvement over state-of-the-art baselines.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Approach for Drug-Drug Interaction Prediction: Utilizing Enhanced Graph Convolutional Networks and 3D Chemical Structures

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Li Wang,
  • Yunfan Li,
  • Kunping Hu,
  • Shiwen Sun,
  • Yingjie Guo

摘要

Accurate drug-drug interaction (DDI) prediction is crucial, yet current graph neural network (GNN) methods face limitations. Their fixed propagation and aggregation schemes struggle to differentiate structurally dissimilar multisets sharing elements, impairing feature extraction and capturing complex graph topologies. Additionally, relying solely on 2D chemical structures fails to reveal complex biological activities or link molecules with different structures but similar effects, limiting prediction accuracy. To overcome these challenges, we propose an improved GCN model incorporating 3D molecular similarity. Our approach innovatively uses a message-passing and aggregation scheme based on neighborhood overlap similarity and Pearson correlation coefficient. We also integrate 3D molecular similarity features during feature fusion. This enhanced model better captures complex drug interaction relationships, achieving a 3% to 16% accuracy improvement over state-of-the-art baselines.