Research Article

Construction of a gene-gene interaction network with a combined score across multiple approaches

Published: June 26, 2015
Genet. Mol. Res. 14 (2) : 7018-7030 DOI: https://doi.org/10.4238/2015.June.26.11
Cite this Article:
A.M. Zhang, H. Song, Y.H. Shen, Y. Liu (2015). Construction of a gene-gene interaction network with a combined score across multiple approaches. Genet. Mol. Res. 14(2): 7018-7030. https://doi.org/10.4238/2015.June.26.11
2,495 views

Abstract

Recent progress in computational methods for inves­tigating physical and functional gene interactions has provided new insights into the complexity of biological processes. An essential part of these methods is presented visually in the form of gene interaction networks that can be valuable in exploring the mechanisms of disease. Here, a combined network based on gene pairs with an extra layer of re­liability was constructed after converting and combining the gene pair scores using a novel algorithm across multiple approaches. Four groups of kidney cancer data sets from ArrayExpress were downloaded and analyzed to identify differentially expressed genes using a rank prod­ucts analysis tool. Gene co-expression network, protein-protein interac­tion, co-occurrence network and a combined network were constructed using empirical Bayesian meta-analysis approach, Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database, an odds ratio formula of the cBioPortal for Cancer Genomics and a novel rank algorithm with combined score, respectively. The topological features of these networks were then compared to evaluate their performances. The results indicated that the gene pairs and their relationship rank­ings were not uniform. The values of topological parameters, such as clustering coefficient and the fitting coefficient R2 of interaction net­work constructed using our ranked based combination score, were much greater than the other networks. The combined network had a classic small world property which transferred information quickly and displayed great resilience to the dysfunction of low-degree hubs with high-clustering and short average path length. It also followed distinct­ly a scale-free network with a higher reliability.

Recent progress in computational methods for inves­tigating physical and functional gene interactions has provided new insights into the complexity of biological processes. An essential part of these methods is presented visually in the form of gene interaction networks that can be valuable in exploring the mechanisms of disease. Here, a combined network based on gene pairs with an extra layer of re­liability was constructed after converting and combining the gene pair scores using a novel algorithm across multiple approaches. Four groups of kidney cancer data sets from ArrayExpress were downloaded and analyzed to identify differentially expressed genes using a rank prod­ucts analysis tool. Gene co-expression network, protein-protein interac­tion, co-occurrence network and a combined network were constructed using empirical Bayesian meta-analysis approach, Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database, an odds ratio formula of the cBioPortal for Cancer Genomics and a novel rank algorithm with combined score, respectively. The topological features of these networks were then compared to evaluate their performances. The results indicated that the gene pairs and their relationship rank­ings were not uniform. The values of topological parameters, such as clustering coefficient and the fitting coefficient R2 of interaction net­work constructed using our ranked based combination score, were much greater than the other networks. The combined network had a classic small world property which transferred information quickly and displayed great resilience to the dysfunction of low-degree hubs with high-clustering and short average path length. It also followed distinct­ly a scale-free network with a higher reliability.

About the Authors