DeepMHAttGRU-DTI: Prediction of Drug-Target Interactions Based on Knowledge Graph Random Walk Embeddings and GRU Neural Network

Recognizing Drug-Target Interactions (DTI) is a crucial step in drug discovery and drug repositioning. Utilizing computational approaches for drug repositioning can reduce experimental costs and expedite drug development. In this paper, a knowledge graph is employed to integrate biological data from various database sources. Subsequently, a graph embedding algorithm based on random walks is utilized to obtain feature vectors of entities in the graph. Following that, data matching is performed on the universal dataset, where successfully matched data is collected and used to construct a binary classification training dataset for deep learning neural network models. The constructed neural network is then trained using drug-target pairs as inputs. Experimental results demonstrate that the node feature vectors obtained using the Monte Carlo Random Walk based on Metropolis-Hastings algorithm (MHRW) based graph embedding algorithm are superior, and the GRU neural network model incorporating multi-head attention mechanism outperforms others.