A Joint Framework for Predicting Disease-Gene Interactions Based on Pre-trained Models and Graph Attention Networks

The study of disease-gene interactions is crucial in biomedical research. Identifying genes associated with diseases can provide critical insights into disease mechanisms, facilitate early diagnosis, and contribute to the development of targeted therapies. In this paper, we propose a novel framework for predicting disease-gene interactions called the PRGAT-DG, which utilizes pre-trained language models and graph attention networks to extract semantic and graph structure features respectively. Moreover, we introduce residual structure to alleviate the problem of excessive smoothing. Experimental results on a dataset released by Stanford University demonstrate the remarkable predictive accuracy of our framework, showcasing its superiority compared to other existing methods. This research holds significant implications for advancing our understanding of disease-gene interaction mechanisms and accelerating the development of relevant therapeutics.