ComPuDTIK: Computational Prediction of Drug-Target Interactions in Kinases

Drug discovery is an intricate and expensive process that involves studying interactions between chemical compounds and protein targets, which are essential for genomic drug discovery and drug repurposing. This study aims to explore diverse computational methods for predicting drug-target binding affinity using the Chembl dataset, containing extensive information on chemical compounds and their interactions with Kinase proteins (Cancer targets). Single-task models, including Random forest, Lasso regression, and Multi-layer perceptron models, are evaluated by creating separate models for each drug-protein interaction. Additionally, multi-task learning is employed using neural networks with task-specific layers to predict binding affinities for multiple interactions simultaneously. The selected optimal model is then used for drug repurposing, analyzing existing drugs for potential use in targeting new conditions. Thus, computational predictions can aid in prioritizing potential drug candidates, and remains crucial before proceeding with clinical trials or treatments.