Drugs–Protein affinity‐score prediction using deep convolutional neural network

Abstract

Abstract Drug discovery involves identifying novel drug–target (DT) interactions. Most proposed computer models for predicting drug–target interactions have emphasized binary classification, but the aim is to determine whether two drug targets interact. However, it is more practical but more challenging to anticipate the binding affinity, which evaluates the strength of a DT pair's association. The drug may not work if the binding affinity is not strong enough. Due to this reason, we need an expert system for predicting the affinity score between the drug and target protein. Advanced deep learning techniques can predict binding affinities because there are more new public affinity data in databases related to DT. This paper uses a comparative analysis of different drug and protein‐encoding techniques to predict DT binding affinities based on similarities between drugs and proteins. The validation results on the standard dataset show that the proposed model is an excellent way to predict how well DT binds and can be very helpful in the process of new drugs. Hence, the model on the DAVIS dataset achieved a higher concordance index, that is, 0.897, and the lowest mean square error, that is, 0.226; for the KIBA dataset, the concordance index score achieved is 0.867, and the mean square error is 0.191. The findings are compared to baseline methods using some evaluation parameters, including the mean squared error and the concordance index.