Abstract
Abstract
Analyzing transcription data requires intensive statistical analysis to obtain useful biological information and knowledge. A significant portion of this data is affected by random noise or even noise intrinsic to the modeling of the experiment. Without robust treatment, the data might not be explored thoroughly, and incorrect conclusions could be drawn. Examining the correlation between gene expression profiles is one way bioinformaticians extract information from transcriptomic experiments. However, the correlation measurements traditionally used have worrisome shortcomings that need to be addressed. This paper compares five already published and experimented-with correlation measurements to the newly developed coincidence index, a similarity measurement that combines Jaccard and interiority indexes and generalizes them to be applied to vectors containing real values. We used microarray and RNA-Seq data from the archaeon Halobacterium salinarum and the bacterium Escherichia coli, respectively, to evaluate the capacity of each correlation/similarity measurement. The utilized method explores the co-expressed metabolic pathways by measuring the correlations between the expression levels of enzymes that share metabolites, represented in the form of a weighted graph. It then searches for local maxima in this graph using a simulated annealing algorithm. We demonstrate that the coincidence index extracts larger, more comprehensive, and more statistically significant pathways for microarray experiments. In RNA-Seq experiments, the results are more limited, but the coincidence index managed the largest percentage of significant components in the graph.