iEdgeDTA: integrated edge information and 1D graph convolutional neural networks for binding affinity prediction

Drug repurposing, where an existing low-risk drug is applied for new indications, becomes more attractive in drug development as drug discovery is very costly and time-consuming. However, the wet-lab testing process to find a drug candidate for a new purpose from its possible binding to a protein is still expensive and laborious due to the available vast quantity of drugs and target proteins. This study aims to leverage artificial intelligence to aid drug repurposing by utilizing drug-protein interaction data and estimating their binding affinity. In this work, we propose an estimation approach that employs a graph-based deep learning technique and enhances prediction accuracy by incorporating the compound's edge information as a multi-dimensional feature. In addition, we used a pre-trained model for protein embedding and graph operation over a 1D protein sequence to overcome a fixed-length problem in the language model task and also incorporated the global feature of the protein. We evaluated the performance of our model in the same benchmark datasets using a variety of matrices, and the results show that our model can achieve the best prediction result compared to other state-of-the-art models while at the same time requiring no contact-map information compared to recent graph-based works Availability: https://github.com/cucpbioinfo/iEdgeDTA