A density-based approach for detecting complexes in weighted PPI networks by semantic similarity.

Hongfang Zhou,Wencong Duan,Junhuai Li,Jie Liu,Byung-Jun Yoon

doi:10.1371/journal.pone.0180570

Hongfang Zhou, Wencong Duan + Show 3 more

Open Access

https://doi.org/10.1371/journal.pone.0180570

Copy DOI

Journal: PloS one	Publication Date: Jul 12, 2017
Citations: 12	License type: CC BY 4.0

Affiliation: University of Technology, Xi'an University of Technology

Abstract

Protein complex detection in PPI networks plays an important role in analyzing biological processes. A new algorithm-DBGPWN-is proposed for predicting complexes in PPI networks. Firstly, a method based on gene ontology is used to measure semantic similarities between interacted proteins, and the similarity values are used as their weights. Then, a density-based graph partitioning algorithm is developed to find clusters in the weighted PPI networks, and the identified ones are considered to be dense and similar. Experimental results demonstrate that our approach achieves good performance as compared with such algorithms as MCL, CMC, MCODE, RNSC, CORE, ClusterOne and FGN.

Highlights

Empirical studies and theoretical modeling of networks have been studied for many years, and some relevant techniques have been improved [1]
The complex biological system that is composed of proteins and Protein-Protein Interaction networks can be described formally as an undirected graph
We put forward a new concept called as Unit Similarity Measure, in which guanosine triphosphate binding (GO) terms annotating proteins are regarded as a semantic collection, and their corresponding DAGs are merged into one united DAG (Directed Acyclic Graph)

Summary

Introduction

Empirical studies and theoretical modeling of networks have been studied for many years, and some relevant techniques have been improved [1]. Experimental results indicate that GMFTP can effectively identify overlapping protein complexes in PPI networks. RFC [24] is a fuzzy clustering algorithm, in which it establishes the fuzzy relationship between proteins and transforms it into some certain equivalence relation The existing methods based on the topology of protein-protein interaction network and biological information have inspired us to improve the accuracy of protein complex recognition. In order to solve the problem caused by false positive and false negative data effectively, we proposed a new method-DBGPWN, which combined the density and semantic similarity in PPI networks. A new concept of semantic clustering coefficient is proposed, which is used for detecting protein complexes in the PPI networks.

Material and method Semantic similarity

Results and discussion

Evaluation metrics