Determination of sets of covariating gene expression using graph analysis on pairwise expression ratios.

Abstract

RNA quantification experiments result in compositional data, however usual methods for compositional data analysis [additive log ratio (alr), centered log ratio (clr), isometric log ratio (ilr)] do not apply easily and give results difficult to interpret. To handle this, a method based on disjoint subgraphs in a graph whose nodes are the quantified RNAs is proposed. Edges in the graph are defined by lack of change in ratios of the corresponding RNAs between conditions. The methods is suited for qRT-PCR and RNA-Seq data analyses, and leads to easy-to-interpret, graphical results and the identification of set of genes that share a similar behavior when the studied condition changes. For qRT-PCR data, it has better statistical properties than the common ΔΔCq method. Construction of all pairwise ratio analysis P-values matrix, and conversion into a graph was implemented in an R package, named SARP.compo. It is freely available for download on the CRAN repository. Example R script using the package are provided as Supplementary Material; the R package includes the data needed. One of these scripts reproduces the Figure2 of this paper. Supplementary data are available at Bioinformatics online.