Whilst the derived transcript abundances from Kallisto can be aggregated to gene level (via tximport R package) and follow DESeq2 analysis, the "natural" downstream processing pipeline of Kallisto results involves Sleuth.
The reason is that Sleuth exploits Kallisto bootstraps to estimate technical noise, when this (technical noise) deviates from the Poisson distribution as is the case of ambiguous origin of the transcripts. Thus, it can efficiently perform differential expression at the transcript level, highlighting biologically important details that are obscured at the gene level (i.e splice variants).
Sleuth lies at the algorithmic framework of general linear models with shrinkage - the concept of gathering information across all transcripts to stabilize variance estimates). Briefly, the key point is that the transcript abundance (observed) is decomposed to the true abundance (unobserved) plus a parameter that accounts for the technical noise, for which bootstraps act as as accurate proxy. The separate modelling of technical and biological variance provides more robust estimates of trancript differential expression.
Programmatically, Sleuth is implemented as an R package along with a Shiny web application allowing user-friendly visualization of the fittted models, MA plots, summaries of the data as well as scatterplots-boxplots to perform diagnostics.
The reason is that Sleuth exploits Kallisto bootstraps to estimate technical noise, when this (technical noise) deviates from the Poisson distribution as is the case of ambiguous origin of the transcripts. Thus, it can efficiently perform differential expression at the transcript level, highlighting biologically important details that are obscured at the gene level (i.e splice variants).
Sleuth lies at the algorithmic framework of general linear models with shrinkage - the concept of gathering information across all transcripts to stabilize variance estimates). Briefly, the key point is that the transcript abundance (observed) is decomposed to the true abundance (unobserved) plus a parameter that accounts for the technical noise, for which bootstraps act as as accurate proxy. The separate modelling of technical and biological variance provides more robust estimates of trancript differential expression.
Programmatically, Sleuth is implemented as an R package along with a Shiny web application allowing user-friendly visualization of the fittted models, MA plots, summaries of the data as well as scatterplots-boxplots to perform diagnostics.
Δεν υπάρχουν σχόλια:
Δημοσίευση σχολίου