A Novel Core‐Attachment–Based Method to Identify Dynamic Protein Complexes Based on Gene Expression Profiles and PPI Networks

Abstract

Cellular functions are always performed by protein complexes. At present, many approaches have been proposed to identify protein complexes from protein–protein interaction (PPI) networks. Some approaches focus on detecting local dense subgraphs in PPI networks which are regarded as protein‐complex cores, then identify protein complexes by including local neighbors. However, from gene expression profiles at different time points or tissues it is known that proteins are dynamic. Therefore, identifying dynamic protein complexes should become very important and meaningful. In this study, a novel core‐attachment–based method named CO‐DPC to detect dynamic protein complexes is presented. First, CO‐DPC selects active proteins according to gene expression profiles and the 3‐sigma principle, and constructs dynamic PPI networks based on the co‐expression principle and PPI networks. Second, CO‐DPC detects local dense subgraphs as the cores of protein complexes and then attach close neighbors of these cores to form protein complexes. In order to evaluate the method, the method and the existing algorithms are applied to yeast PPI networks. The experimental results show that CO‐DPC performs much better than the existing methods. In addition, the identified dynamic protein complexes can match very well and thus become more meaningful for future biological study.

A Special Structural Based Weighted Network Approach for the Analysis of Protein Complexes

“…We discovered on the one hand that the majority of algorithms only consider the percentage of detected complexes. The p-values of identified protein complexes, on the other hand, are proportional to their size [22,23,84,85]. When analyzing the function enrichment of identified protein complexes, it is essential to consider both the quantity and the proportion of the identified complexes.…”

Spatiotemporal constrained RNA–protein heterogeneous network for protein complex identification

“…GHAE [ 6 ] integrates gene ontology attributes in heterogeneous networks, learns protein embeddings through attention mechanisms and screens core proteins and attachments based on embedding similarity. CO-DPC [ 20 ] screens active proteins based on gene expression profile information with the help of the 3-sigma principle and constructs dynamic PPI networks. CO-DPC identifies locally dense subgraphs as protein cores and further complements the attachments.…”

“…Existing methods, while attempting to utilize the structural and other biological information of PINs, are limited by the expressive power of homogeneous PINs and the homogenization of the biological information to be learned [ 20 , 25 ]. This further affects the biological significance of protein complex identification.…”