Identification of essential proteins based on a new combination of topological and biological features in weighted protein–protein interaction networks

Abstract

The identification of essential proteins in protein-protein interaction (PPI) networks is not only important in understanding the process of cellular life but also useful in diagnosis and drug design. The network topology-based centrality measures are sensitive to noise of network. Moreover, these measures cannot detect low-connectivity essential proteins. The authors have proposed a new method using a combination of topological centrality measures and biological features based on statistical analyses of essential proteins and protein complexes. With incomplete PPI networks, they face the challenge of falsepositive interactions. To remove these interactions, the PPI networks are weighted by gene ontology. Furthermore, they use a combination of classifiers, including the newly proposed measures and traditional weighted centrality measures, to improve the precision of identification. This combination is evaluated using the logistic regression model in terms of significance levels. The proposed method has been implemented and compared to both previous and more recent efficient computational methods using six statistical standards. The results show that the proposed method is more precise in identifying essential proteins than the previous methods. This level of precision was obtained through the use of four different data sets: YHQ-W, YMBD-W, YDIP-W and YMIPS-W.

Network Learning for Biomarker Discovery

“…The degree centrality, information centrality, betweenness centrality and other dependent centralities have been used to predict essential proteins from PPI networks with great accuracy [33,34]. The fundamental dependent centrality and eigenvector centrality have been used to predict essential proteins [34][35][36][37][38]. Tang et al [39] develop a software package to predict essential proteins based on several commonly used centralities, including degree centrality, betweenness centrality, information centrality, and eigenvector centrality, to name a few.…”