Differential variation and expression analysis

Abstract

We propose an empirical Bayes approach using a three-component mixture model, the L2N model, that may be applied to detect both differential expression (mean) and variation. It consists of two log-normal components (L2) for the differentially expressed (dispersed) features (one component for under-expressed [dispersed] and the other for over-expressed [dispersed] features), and a single normal component (N) for the null features (i.e., non-differentially expressed [dispersed] features). Simulation results show that L2N can capture asymmetries in the numbers of over-and under-expressed (dispersed) features (e.g., genes) when they exist, can provide a better fit to data in which the distributions of the null and non-null features are not well-separated, but can also perform well under symmetry and separation. Thus the L2N model is particularly appealing when no a priori biological knowledge about the mixture density is available. The L2N model is implemented in an R package called DVX, for Differential Variation and eXpression analysis. The package also includes an implementation of differential expression analysis via the limma package, and a differential variation and expression analysis using a three-way normal mixture model. DVX is a user-friendly, graphical interface implemented via the Shiny package [6], so that a user is not required to have R programming knowledge. It offers a set of diagnostics plots, data transformation tools, and report generation in Microsoft Excel- and Word-compatible formats. The package is available on the web, at https://haim-bar.uconn.edu/software/DVX/.