Inferring Gene Regulatory Networks Using Kendall's Tau Correlation Coefficient and Identification of Salinity Stress Responsive Genes in Rice

Abstract

Salinity is one of the most common abiotic stresses that limit the production of rice. Since salinity stress tolerance is controlled by many genes, identification of these stress responsive genes as well as to understand the underlying mechanisms is of importance from breeding point of view. In this direction, the reverse engineering of gene regulatory networks has proven to be successful. In this study, we construct the gene regulatory network using Kendall's tau correlation coefficient, in order to identify the stress responsive genes. The proposed approach was tested on a rice microarray dataset and 18 key genes were identified. Most of these key genes were found to be involved directly or indirectly in salinity stress, as evidenced from the published literature. Gene ontology analysis further confirmed the involvement of the selected genes in ion binding, oxidation-reduction and phosphorylation activities. These identified genes can be targeted for breeding salt-tolerant varieties of rice.

Statistical Approach for Biologically Relevant Gene Selection from High-Throughput Gene Expression Data

“…Filter methods select individual genes or gene subset based on a performance measure computed from the data with respect to class variables regardless of the predictive modeling algorithm [ 17 ]. These methods include univariate approaches such as t -test [ 18 , 19 ], Fold change [ 19 ], F-score [ 20 , 21 ], Volcano plot [ 18 ], Wilcoxon’s statistic (Wilcox) [ 22 , 23 ], information gain (IG) [ 9 , 24 ], gain ratio (GR) [ 9 , 24 ], symmetric uncertainty [ 19 ], etc. These methods select genes by only considering their relevance within a level of the experimental condition/trait.…”

“…Through this, relevant genes are selected from a high-dimensional GE data through the statistical significance values computed using a nonparametric (NP) test statistic under a bootstrap-based subject sampling model. Further, the comparative performance analysis of the proposed BSM approach is carried out with nine existing competitive methods (i.e., IG [ 9 , 24 ], GR [ 9 , 24 ], t -test [ 18 , 19 ], F-score [ 20 , 21 ], MRMR [ 12 , 20 ], SVM-RFE [ 8 , 29 ], SVM-MRMR [ 13 ], PCR [ 9 , 24 ] and Wilcox [ 22 , 23 ]). The comparative performance measures include CA along with its standard error computed through varying sliding windows size technique, and three biological criteria based on QTL [ 42 ] and GO [ 43 ] terms.…”

Statistical Approach for Biologically Relevant Gene Selection from High-Throughput Gene Expression Data

“…Filter methods select individual genes or evaluate a gene subset based on a performance measure computed from the data with respect to class variables regardless of the predictive modeling algorithm [14]. Further, these methods include univariate approaches such as t-test [15,16], Fold change [16], F-score [17,18], Volcano plot [15], Wilcoxon's statistic (Wilcox) [19,20], Information Gain (IG) [21,22], Gain Ratio (GR) [21,22], symmetric uncertainty [19], etc. These methods select genes by only considering their relevance within a level of the experimental condition/trait.…”

“…Further, the comparative performance analysis of the proposed BSM approach is carried out with nine existing competitive methods (i.e. IG [21,22], GR) [21,22] , t-test [15,16], F-score [17,18], MRMR [17,39] , SVM-RFE [8,26], SVM-MRMR [28], PCR [21,22] and Wilcox [19,20]). The comparative performance measures include, CA along with its standard error computed through varying sliding windows size technique, and two biological criteria based on QTL [40], and GO [41] terms.…”