Graph-Based Framework for Anti-Cancer Drug-Target Interaction Prediction using Proteomics, Chemical Space and Genomic Features

Drug target interaction (DTI) prediction is a critical step in drug discovery, that helps to accelerate the targeted drug therapy. Different machine learning methods have been utilized to develop in-silico DTI prediction. However, these models mainly focused on drugs and genomics data. This study has focused on the target- level information in the model developed. For this, we have represented the graph neural network (GNN) based framework that integrated the molecules 2D features, protein sequence-based features, amino acid contact map and genomics features including gene expression and copy number variation (CNV) of cancer cell lines from GDSC database. Our results show that among different GNN models developed in this study, graph attention neural network (GAT) performed well with a R² value of 0.677. Primarily, the integration of target representation significantly improves the model learning process compared to the other studies. Previous studies reported RMSE value ranges from 1.2-0.8, our proposed model reduce the error significantly, achieving an RMSE of 0.081 in drug response prediction. Subsequently, the multimodal data fusion and attention mechanism of graph learning provides a biologically informed and scalable approach for precision oncology.