Two Reasons Why DESeq2 Is A Widely Popular Choice in Differential Expression Analysis for Single-cell RNA-Seq #1 Good estimation of variance through the gene-wise dispersion parameter of Negative Binomial distribution For single cell RNA seq data, it is recommended this second test is used over the Wald test. If not, then we are better off with the simpler model ( Occam’s razor ).
#MODERATED ESTIMATION OF FOLD CHANGE AND DISPERSIO FULL#
The basic idea of this test is to see if, by the addition of another variable (the variable of interest), the full model can do a better job fitting to the data at hand. These two models both contain confounding variables that need to be accounted for, yet the more complex one (i.e., the full model) also contains the variable of interest. The Likelihood Ratio Test, on the other hand, aims to compare two models to see which one better explains the observed data. The Wald test is a pairwise comparison where the null hypothesis is that the log fold change (the gene model coefficient) is zero, indicating no significant differential expression between two groups. Here we have two options: the Wald test or the Likelihood Ratio test In particular, the log fold change in the expression of gene i in sample j is calculated from the sum of fold changes related to all design factors taken into accountįinally : run a statistical test to examine if the changes are statistically significant. This is also called the Design Formula, which users must specify before running DESeq2.įormula 3: After fitting, the coefficient ẞ i provides information about the log fold change. While the exact details for the fitting could be found in linked articles at the end of the blog, the end result is coefficients that represent expression change between conditions/samples (i.e log fold changes) (Formula 2, 3).įormula 1 : The read count K ij (for gene i in sample j) is fitted to a Negative Binomial GLM with two parameters mean expression value and dispersion (which we have calculated and estimated in step 1)įormula 2 : An easier interpretation of the Negative Binomial GLM in Formula 1 with y is the gene expression (obtained from the mean and dispersion), x i is the design factor i (e.g control = 0, case = 1), ẞ i is the coefficient for design factor i (explained in detail here ). Second : with the dispersion estimated, the count value of each gene is fitted to a Negative Binomial Generalized Linear Model (Formula 1). Due to the nature of single-cell RNA seq data, the ratio of mean and variance shifts when the expression level of a gene changes rather than remains constant, thus it is important to accurately estimate the relationship between these two statistics via dispersion for subsequent model fitting step. The highlight of this step is the estimation of dispersion, a statistic reflecting the relationship between mean and variance. Here’s an oversimplified description of what DESeq2 does stepwise and what each step means.įirst : calculate the mean and estimate the dispersion of gene expression for each and every gene. What makes DESeq2 a widely popular tool in single-cell analysis? In this article, we’ll break down the basics of DESeq2 and highlight 2 advantages of DESeq2 in differential gene expression for single-cell RNA-seq. This time, we’d like to discuss a frequently used tool – DESeq2 ( Love, Huber, & Anders, 2014).Īccording to Squair et al., (2021), in 500 latest scRNA-seq studies, only 11 methods made up to over 80% of DE analysis. In our previous post, we have given an overview of differential expression analysis tools in single-cell RNA-Seq. Differential expression analysis is a common step in a Single-cell RNA-Seq data analysis workflow.
0 kommentar(er)
