Protein function prediction method and system fusing hits algorithm and gcn

Abstract

The application belongs to the technical field of protein function prediction, and specifically discloses a protein function prediction method and system fusing HITS algorithm and GCN, which comprises the following steps: constructing a heterogeneous biological network and an adjacency matrix of the heterogeneous biological network; running the HITS algorithm on the heterogeneous biological network, and iteratively calculating the authority value and hub value of each protein node and each GO term node based on the adjacency matrix; extracting protein sequence features and GO term features, and fusing them into a joint feature matrix; inputting the joint feature matrix and the adjacency matrix into a multi-layer graph convolution network to learn the local context representation of each node; constructing a supervision signal and a HITS score based on the authority value and the hub value, fusing the supervision signal and the local context representation, and then generating a final protein function prediction result through an iterative propagation mechanism based on the HITS score. The application solves the problem that the prior art is difficult to capture long-range dependence and the importance of global function in a heterogeneous graph.

Images