Identification of differentially distributed gene expression and distinct sets of cancer-related genes identified by changes in expression mean and variability
AbstractBackgroundDifferential expression analysis of RNA-seq data has advanced rapidly since the introduction of the technology, and methods such as edgeR and DESeq2 have become standard parts of analysis pipelines. However, there is a growing body of research showing that differences in variability of gene expression or overall differences in the distribution of expression values – differential distribution – are also important both in normal biology and in diseases including cancer. Genes whose expression differs in distribution without a difference in mean expression level are ignored by differential expression methods.ResultsWe have developed a Bayesian hierarchical model which improves on existing methods for identifying differential dispersion in RNA-seq data, and provides an overall test for differential distribution. We have applied these methods to investigate differential dispersion and distribution in cancer using RNA-seq datasets from The Cancer Genome Atlas. Our results show that differential dispersion and distribution are able to identify cancer-related genes. Further, we find that differential dispersion identifies cancer-related genes that are missed by differential expression analysis, and that differential expression and differential dispersion identify functionally distinct sets of genes.ConclusionThis work highlights the importance of considering changes beyond differences in mean in the analysis of gene expression data, and suggests that analysis of expression variability may provide insights into genetic aspects of cancer that would not be revealed by differential expression analysis alone. For identification of cancer-related genes, differential distribution analysis allows the identification of genes whose expression is disrupted in terms of either mean or variability.