Caltech researchers describe sleuth, a method for the differential analysis of gene expression data that utilizes bootstrapping in conjunction with response error linear modeling to decouple biological variance from inferential variance. sleuth is implemented in an interactive shiny app that utilizes kallisto quantifications and bootstraps for fast and accurate analysis of data from RNA-seq experiments.
Overview of Sleuth
(a) sleuth models different sources of variance to predict differentially expressed transcripts and genes. Biological variance (biol. var.) results from differences in RNA content between replicates and from stochastic biochemistry during library preparation, while inferential variance arises from random sequencing and computational analysis of reads. (b) Results for an example gene after running kallisto on RNA-seq data from Trapnell et al.12 generated from human lung fibroblasts transfected with scrambled siRNA (scramble condition) and HOXA1 siRNA (HOXA1KD condition). DESeq2 and voom identify the gene as differentially expressed, but high inferential variance causes sleuth to find no difference. Red dots, point estimates. Blue dots, results for bootstrap samples to assess inferential variance. (c) The between-sample raw variance leads to a small estimated biological variance that fails to account for uncertainty introduced when quantifying the samples
Availability –Sleuth may be downloaded at: http://pachterlab.github.io/sleuth