Drug-Target Affinity Prediction Based on Improved GraphDTA

The development of new drugs is costly, time-consuming, and high-risk. Drug repurposing is a strategy that can avoid the expensive and lengthy drug development process. To achieve effective drug repurposing, understanding which proteins are targets of which drugs is a crucial issue. Predicting drug-target affinity using computational models can accelerate drug repurposing. However, many existing computational models represent drugs as strings, which are not the optimal representation of molecules. In GraphDTA, the SMIILES encoding of molecules is converted into a two-dimensional molecular graph, and a graph neural network is used to better learn the features of molecules. This article improves the GraphDTA model and proposes two methods to further enhance the model’s performance. Firstly, a multi-scale feature fusion mechanism is introduced in the GIN model, which can better capture the structural information and chemical features of molecules, improving the accuracy of the model. Secondly, we propose a method that combines GIN and GAT. By concatenating the GIN and GAT models, a more comprehensive and accurate graph representation can be obtained, improving the model’s predictive performance. The model is evaluated using the Davis dataset, and the results show that the proposed model outperforms the original model, demonstrating the effectiveness of this method and its potential for a wide range of applications.