Graph Neural Network-based Drug-Target Interaction Prediction for Precision Medicine
Yuhan Dai, Li Wei, Yu Chen · Optimizations in Applied Machine Learning · 2025
Drug-target interaction (DTI) prediction is central to precision medicine because it supports target prioritization, drug repurposing, and patient-specific therapeutic design. Existing computational approaches, however, often struggle with sparse interaction labels, heterogeneous biological evidence, and the nonlinear topology of drug-target networks. This paper proposes a Graph Neural Network (GNN)-based DTI prediction framework that represents drugs and targets as nodes in a bipartite biological graph and learns interaction-aware embeddings through message passing. By integrating drug descriptors, target features, structural similarity, and graph connectivity, the framework captures relationships that are difficult for similarity-based, matrix-factorization, or purely tabular models to express. A case study using simulated biochemical descriptors illustrates the modeling workflow, evaluation metrics, and limitations of baseline prediction. The results emphasize that graph-based representation learning can provide a more flexible foundation for DTI screening and precision-medicine decision support.