Abstract
Summary Single cell transcriptomics provides a window into cell-to-cell variability in complex tissues. Modeling single cell expression is challenging due to high noise levels and technical bias. In the past years, considerable efforts have been made to devise suitable parametric models for single cell expression data. We use Discrete Generalized Beta Distribution (DGBD) to model read counts corresponding to a gene as a function of rank. Use of DGBD yields better overall fit across genes compared to the widely used mixture model comprising Poisson and Negative Binomial density functions. Further, we use Wald’s test to probe into differential expression across cell sub-types. We package our implementation as a standalone software called ROSeq. When applied on real data-sets, ROSeq performed competitively compared to the state of the art methods including MAST, SCDE and ROTS.
Software Availability The Windows, macOS and Linux - compatible softwares are available for download at https://malaalam.github.io/ROSeq
Contact abhianik{at}gmail.com debarka{at}iiitd.ac.in