AGDNGDA: Unraveling Drug-Associated Genes with Adaptive Graph Diffusion Networks

Understanding the intricate relationships between genes and drugs is crucial for advancing drug discovery. However, biological experiments aimed at identifying gene-drug associations are typically time-consuming and inefficient, leading to significant data sparsity. While Graph Neural Networks (GNNs) have demonstrated effectiveness in addressing sparse data, existing approaches often rely on uniform node processing or static attention mechanisms, which limit their ability to capture the dynamic nature of biological interactions. To overcome these limitations, we propose AGDNGDA, an adaptive graph diffusion network that leverages a heat kernel mechanism to dynamically modulate information aggregation according to each node's specific context. This innovative approach enables our model to effectively highlight significant yet challenging gene-drug interactions. Comprehensive experimental results demonstrate that AGDNGDA consistently surpasses existing state-of-the-art methods. Additionally, detailed case studies illustrate the model's capability in identifying biologically meaningful gene-drug associations, highlighting its potential as a powerful tool for pharmaceutical research.