BiProSimGCN: Prediction of Drugs-Proteins Interactions Using Graph Convolutional Neural Network and Proteins Similarity

The drug discovery process faces challenges due to its length, cost, and experimental failures. Researchers are addressing these issues through “drug repurposing,” seeking new applications for existing pharmaceuticals. To predict drug-target interactions, computational methods, like drug-target interaction prediction, utilize bipartite graphs and link prediction. In this paper, we propose BiProSimGCN, a framework based on convolutional graph neural networks, to predict drug-protein interactions. BiProSimGCN integrates data from the bipartite drug-protein network and the protein similarity network, aiming to reveal latent factors of drugs and proteins. The performance of BiProSimGCN is evaluated using a dataset with information about drugs, target proteins, and their interactions, as well as benchmark datasets with interactions involving enzymes, ion channels, GPCRs, and nuclear receptors. The results indicate that BiProSimGCN outperforms several models from the literature and achieves comparable performance with others.