Bayesian approach to single-cell differential expression analysis

Abstract

Single-cell data provides means to dissect the composition of complex tissues and specialized cellular environments. However, the analysis of such measurements is complicated by high levels of technical noise and intrinsic biological variability. We describe a probabilistic model of expression magnitude distortions typical of single-cell RNA sequencing measurements, which enables detection of differential expression signatures and identification of subpopulations of cells in a way that is more tolerant of noise.

Opportunities and Challenges in Implementation of Multiparameter Single Cell Analysis Platforms for Clinical Translation

“…A number of approaches are being considered for assuring confidence in these measurements, including the use of the External RNA Controls Consortium (ERCC) spike‐in material34, 35 to provide a metric of accuracy for those known sequences. Other approaches include the use of Bayesian statistics to assess real differences in the presence of dropout events36 that occur due to the low amount of RNA in cell samples and result in detection of the gene in some cells and not in others. When the apparent heterogeneity in gene expression is simply due to technical issues, the data can lead to erroneous conclusions of biological heterogeneity.…”

“…Multiplex protein expression (4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37) …”

DEsingle for detecting three types of differential expression in single-cell RNA-seq data

“…As a result, mRNA molecules in a cell can be randomly missed during the reverse transcription step and the following cDNA amplification step, and the mRNA products of some genes may be totally missed in the capturing procedure, which then produces dropout zeros in the scRNA-seq data (3,25,26). In this section, we try to model this mRNA capture procedure and study what impact this process will have on the ZINB distribution.…”

“…This phenomenon is called dropout events (25,26). We call this type of zero values as dropout zeros.…”

“…The methods we compared with include traditional DE detection methods edgeR (16) and DEGseq (17) that were developed for bulk RNA-seq data but have been also used on many scRNA-seq data (12), as well as new methods specifically developed for scRNA-seq data including BPSC (34), D3E (19), monocle (35), SCDE (25). We applied DEsingle on a real scRNA-seq dataset of human preimplantation embryonic cells of different days of embryo development (36) and found interesting and 10% represent the observed data obtained from the original data after mRNA capture procedure.…”