Abstract
Drug-target affinity (DTA) prediction is a fundamental challenge in drug discovery. Computational methods for predicting DTA can greatly assist drug design by narrowing the search space and reducing the number of protein-ligand complexes with low affinity. Currently DTA approaches often do not require three-dimensional (3D) structural information of proteins, which is frequently unavailable. In this study we present the DrugForm-DTA model, which uses only structure-less representations of ligand and protein. It is a Transformer-based neural network with protein encoding based on ESM-2, and small molecule ligand encoding obtained with Chemformer. We evaluated the model on the standard benchmarks Davis and KIBA, and revealed superior performance of DrugForm-DTA with the best result for KIBA. Moreover, we developed a ready-to-use model trained on the BindingDB dataset which was subjected to high-quality filtering and transformation. Overall, our method predicts drug-target affinity values with a confidence level comparable to that of a single in vitro experiment. Also, we compared DrugForm-DTA against molecular modeling methods and revealed higher efficacy of the developed model for drug-target affinity predictions. Our investigation provides a high accuracy neural network model with performance comparable to that of experimental measurements, a filtered.and reassessed BindingDB dataset for further usage, and demonstrates the outstanding applicability of the proposed method for DTA prediction.