GATPDD: an Enhanced Deep Learning Framework for Predicting Drug-Parasitic Disease Associations.

Abstract

Parasitic diseases pose a significant threat to human health. Accurate and robust prediction of drug-parasitic disease associations is critical to advancing drug discovery and developing parasitic disease therapies. However, biomedical data in this field is often too scarce to train a generalized machine learning model. Although computational methods have been developed for predicting potential drug-parasitic disease associations, their performances were restricted owing to data limitation. Here we propose a deep learning framework entitled GATPDD for improving drug-parasitic disease association predictions. Our model integrates enhanced Deep Graph Infomax with multi-head Graph Attention Networks and Neighborhood Interaction Attention to refine feature learning and embedding aggregation in the scenario of limited benchmark datasets. Extensive comparative experiments demonstrate that GATPDD effectively alleviates the data scarcity problem for the model generalization and significantly improves accuracy and robustness over state-of-the-art methods. We further use GATPDD to conduct case studies and results validate its ability to identify reliable drug-parasitic disease associations in real-world applications, suggesting the potential of GATPDD in drug discovery for parasitic disease therapies.