An Integrative Deep Learning Model for Drug-Target Interaction Prediction

Abstract

The technological advancements in modern therapeutic studies enable progressively more effective approaches to disease treatments. Despite the investigation for drug interactions remaining crucial, it is also very expensive and time-consuming. Considering especially this second aspect, several strategies have been recently proposed to accelerate the discovery of a treatment solution. The following paper introduces a model for drug-target interaction prediction. The architecture is conceived to generate feature vectors encoding for both drugs and targets, using specifically designed neural networks. Further data processing works on the two entities separately, combining them just before the final prediction. The model calculates a Boolean output that implies the possible presence, or absence, of a beneficial reaction for the patient. The experiment and the comparisons described in this work also prove the strengths of our method and establish a promising starting point for future developments.