Affiliations: Department of Computer Science and Engineering, Netaji Subhash Engineering College, Kolkata, India | Machine Intelligence Unit, Indian Statistical Institute, Kolkata, India
Note: [] Corresponding author. Rajat K. De, Machine Intelligence Unit, Indian Statistical Institute, Kolkata 700108, India. E-mails: [email protected] (Rajat K. De); [email protected] (Anupam Ghosh).
Abstract: In this article, we propose a methodology for identifying the interactions among the genes in terms of dependencies (named as gene–gene interaction) that have altered quite significantly from normal stage to diseased stage with respect to their expression patterns. This idea leads to predict the disease mediating genes along with their altered interactions. The proposed methodology involves measuring information content of individual genes using fuzzy entropy, conditional fuzzy entropy of a gene on another, dependencies (interactions) of a pair of genes in both normal and diseased states, detecting the dependencies being deviated from normal to carcinogenic state and finally identifying the influential genes from altered dependencies. Thus the gene–gene interactions for normal state and diseased state are represented separately by the gene dependency networks (GDN). The altered interactions among the genes have been represented using a network, called altered gene dependency network (AGDN), in which each node represents a gene and a directed edge signifies altered dependency between a pair of nodes (genes). The methodology has been demonstrated on five gene expression data sets dealing with human lung cancer, colon cancer, sarcoma, breast cancer and leukemia. The results are appropriately validated, in terms of gene–gene interactions, using biochemical pathways, t-test, p-value, NCBI database and earlier investigations in terms of gene regulation. We have also used sensitivity to validate the results. For a comparative study, we have used some existing association rule mining algorithms and frequent pattern mining algorithms like Fuzzy Cluster-Based Association Rules, Apriori, T-Apriori in terms of gene–gene interactions. In addition, we have implemented Significance Analysis of Microarray, Signal-to-Noise Ratio, Neighborhood analysis, Bayesian regularization and frequent pattern mining algorithms for a comparison with AGDN in terms of ability to identify the important genes mediating the cancers.
Keywords: Gene dependency networks, altered gene dependency networks, lung cancer, colon cancer leukemia, sarcoma, breast cancer, t-test, p-value
