Inference of Gene Regulatory Networks Using Coefficient of Determination, Tsallis Entropy and Biological Prior Knowledge

In this work, we studied a common problem in Systems Biology, which is the inference or reverse engineering of gene regulatory networks from gene expression data. We addressed this problem using the Boolean formalism, where the expression of a gene is represented only by two possible values: 0 (not expressed) or 1 (expressed). Besides that, our methodology is based on a feature selection approach and we used an algorithm named IFFS - improved forward floating selection. We performed experiments to compare two measures of gene interactions used in the criterion function of the algorithm, the coefficient of determination and the mutual information computed via Tsallis entropy. Besides that, we also incorporated a biological prior knowledge source of gene interactions from a database known as STRING. To validate the methodology, we used data from the DREAM challenge and a dataset from a budding yeast cell-cycle study. The results showed that, generally, the mutual information performs slightly better than the coefficient of determination, and that incorporating biological knowledge improves the results.