Abstract
Understanding gene-drug associations is essential in drug discovery, where advances in artificial intelligence and data-driven methods have revolutionized the identification of novel therapeutic applications, molecular pathways, and potential drug targets for existing medications. However, current computational methods are hindered by data sparsity and limited capacity to model the complex interactions between genes and drugs. To address these challenges, we propose a novel computational framework-disentangled contrastive learning with dynamic intent adaptation (DIACL)-for predicting unknown gene-drug associations. DIACL leverages disentangled contrastive learning to decompose the latent factors driving drug-gene interactions, yielding more robust and interpretable feature representations. Additionally, we introduce a dynamic intent representation mechanism and an adaptation graph augmentation strategy to enhance the model's ability to capture fine-grained interaction details. Extensive experiments on benchmark datasets demonstrate that DIACL significantly outperforms state-of-the-art methods in terms of prediction accuracy and generalization capability. Our findings highlight DIACL's potential as a scalable and efficient tool for accelerating drug discovery and advancing precision medicine by identifying therapeutic targets.