A Novel Method to Predict Drug-Target Interactions Based on Large-Scale Graph Representation Learning

Abstract

Identification of drug-target interactions (DTIs) is a significant step in the drug discovery or repositioning process. Compared with the time-consuming and labor-intensive in vivo experimental methods, the computational models can provide high-quality DTI candidates in an instant. In this study, we propose a novel method called LGDTI to predict DTIs based on large-scale graph representation learning. LGDTI can capture the local and global structural information of the graph. Specifically, the first-order neighbor information of nodes can be aggregated by the graph convolutional network (GCN); on the other hand, the high-order neighbor information of nodes can be learned by the graph embedding method called DeepWalk. Finally, the two kinds of feature are fed into the random forest classifier to train and predict potential DTIs. The results show that our method obtained area under the receiver operating characteristic curve (AUROC) of 0.9455 and area under the precision-recall curve (AUPR) of 0.9491 under 5-fold cross-validation. Moreover, we compare the presented method with some existing state-of-the-art methods. These results imply that LGDTI can efficiently and robustly capture undiscovered DTIs. Moreover, the proposed model is expected to bring new inspiration and provide novel perspectives to relevant researchers.

Keywords: computational method; drug discovery; drug-target interactions; large-scale graph representation learning.

Figure 1

A schematic diagram of the drug molecular structure is constructed as bit vectors. A is the structure of a drug molecule, and B, C, and D are all substructures of the drug molecule, corresponding to the converted bit (represented by the small black box), respectively.

Figure 2

An example of large-scale graph representation learning. (A) The schematic diagram of the relationship between drugs and targets. (B) An example of the graph embedding in drug-target interactions (DTIs). (C) An example of the graph convolutional network.

Figure 3

The flowchart of the proposed large-scale graph representation learning DTI (LGDTI). (a) A bipartite graph of DTIs. The solid black line is described as known DTIs, and the dashed red line is described as latent DTIs. (b) Part A constructed an adjacency graph containing a self-loop, in which green nodes are drugs and purple nodes are targets, and the information of first-order neighbors of each node is aggregated through graph convolutional network. Part B represented high-order information of each node in a bipartite graph by DeepWalk. (c) The two kinds of representation features are integrated. (d) Random forest classifier is trained and used for predicting new DTIs.

Figure 4

The receiver operating characteristic (ROC) and precision-recall (PR) curves under 5-fold cross-validation.

Figure 5

Comparison of the ROC and PR curves performed based on different machine learning classifier.

Figure 6

Comparison of the ROC and PR curves performed by random forest classifier based on different features.