Integrating graph structure information and node attributes to predict protein-protein interactions

Abstract

Protein-protein interactions (PPIs) greatly influence cellular biological processes. Therefore, identifying protein-protein interactions shed valuable light on the structure of life activities. Different computational methods for PPI detection have recently been proposed, but most of them have used either node attributes or graph structure information. Meanwhile, it has been found that integrating node attributes into graph structure information could be potentially helpful in other complex networks. In the present research, a graph embedding method is suggested to incorporate these two kinds of information for the efficiency of several downstream tasks such as predicting interaction between proteins. At first, an enriched denser graph is reconstructed with graph structure information and protein evolutionary information. Then, a biased random walk model is used to produce sequences of proteins, and finally these sequences are given to skip-gram with negative sampling model to learn the low-dimensional representation of each protein. The proposed method is evaluated on several real-world datasets. The outcomes present the efficiency and effectiveness of the suggested method in comparison with the other methods.