This paper is published in… By Zhou Yaoqi research group of Shenzhen Graduate School of Peking University and Yang Yuedong research group of Sun Yat sen University Bioinformatics Articles on .
Address of thesis ：https://doi.org/10.1093/bioinformatics/btab643
Abstract
protein – Protein interactions （PPI） It plays a vital role in many biological processes , distinguish PPI Site is an important step in understanding disease mechanism and designing new drugs . because PPI The experimental method of site recognition is expensive and time-consuming , Many computational methods have been developed as screening tools . However , Most of these methods are based on the adjacent features in the sequence, and are limited to obtaining spatial information .
The author proposes a method for PPI A depth map based framework for site prediction GraphPPIS（ Depth map convolution network for protein interaction site prediction ）, among PPI The locus prediction problem is transformed into a graph node classification task , And by using the initial residual (initial residual) Identity mapping (identity mapping) Deep learning of technology to solve . The author shows , Compared with other sequence based and structure based methods , stay AUPRC and MCC On , Deeper Architecture （ As many as 8 layer ） The performance can be improved respectively 12.5% and 10.5% above . Further analysis shows that , Even if a false positive prediction is made ,GraphPPIS The predicted interaction sites are more spatially clustered , Closer to the native site . The results highlight the importance of capturing spatially adjacent residues for the prediction of interaction sites .
GraphPPIS Network framework of model
Read more here: Source link