K4. Gene network construction and pathways analysis for high throughput microarrays

Abstract

The key idea discussed in this paper is to infer gene regulatory network from high throughput microarray data for Hepatocellular Carcinoma (HCC). Working with such huge number of genes is a complex process. So, our framework for inferring gene interactions from large scale microarrays is based on a selected set of informative genes. We applied two measures of dependencies between genes: Correlation and mutual information. Therefore, two types of networks were constructed: Co-expression network and Mutual information network. Some Mutual information network inference algorithms: Context Likelihood of Relatedness (CLR), Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE), and Minimum Redundancy Network (MRNET) were applied. A proposed method for simplifying the complex structure of the inferred network is introduced using the Minimum Spanning Tree (MST) which provides a better visual interpretation of the constructed networks. From the constructed networks we were able to identify a set of functional gene modules. These modules were validated using the Gene Ontology (GO) enrichment. The GO enrichment analysis has proven the strength of the ARACNE inference algorithm over all other employed algorithms. Moreover, a comparison was carried out between the Mutual information network inference and the well known Bayesian inference. To establish this comparison, specific pathways in HCC were rather chosen. These pathways were tested for their significance using singular value decomposition. According to this comparison, again the ARACNE showed better results.