A Prediction Model of Drug-Drug Interactions Based on Graph Neural Networks and Subgraph Attention Mechanisms

This paper introduces a novel model for predicting Drug-Drug Interactions (DDIs) by utilizing a Self-Attention Graphical Neural Network (SAGNN) mechanism. DDIs are recognized for inducing unpredictable pharmacological reactions, and previous studies have struggled to identify their causal mechanisms. Moreover, Graph neural networks (GNNs) have demonstrated exceptional performance in computational chemistry and have been widely employed for mining DDIs. The accurate identification of main functional groups within drug compound molecules is crucial for predicting DDIs, presenting a specific challenge for GNNs. To address this issue, the SAGNN model is proposed in this work. The SAGNN model comprises a linear modulation feature, an attention mechanism module, and a topological pooling module with an added subgraph attention mechanism. The former module is capable of adaptively perceiving various functional group structures based on different drug compound molecule configurations, while the latter module computes the interactions between the primary functional groups in drug compound molecules using the added subgraph attention mechanism. Through the analysis of experimental results, it has been shown that the SAGNN model accurately identifies functional group structures within drug compound molecules. These advantages underscore the potential of the proposed model to substantially enhance DDI prediction capability.