DTE and DTU without biological replicates ?

[med1104]

In the example "Differential Transcript Usage on a non model organism", the script for DTE analysis requires having a biological replicate. Is it possible do the analysis without biological replicate? The DTE script of the example (voom_diff.R) doesn't run if not having biological replicate.

I modified the script and now it runs without biological replicates, but iI don't know if the analysis is correct.

My script is this:

# library
library('edgeR')

## Read in data
counts <- read.table("Counts/counts.txt",header=TRUE,sep="\t")

##Define groups
treatment = c('T1','T2')

## Make exon id
eid = paste0(counts$Chr,"-S",counts$Start,"-E",counts$End)

## Define design matrix
design <- model.matrix(~treatment)

## Make DGElist and normalise
dx <- DGEList(counts[,c(7:8)])
dx <-calcNormFactors(dx,group= treatment )

## glmFit
gfit <- glmFit(dx, design, dispersion = 0.1)
ds <- diffSpliceDGE(gfit, geneid = counts$Chr, exonid = eid)

## Results
topSpliceDGE(ds, number = 20, test = "Simes")
plotSpliceDGE(ds)

[Quarkins]

Hi,
Thanks for your interest in the Lace software!
Generally you usually want at least a couple of biological replicates in order to improve your statistical strength and uncertainties when comparing two different treatments like in this case. One can always use just one replicate per treatment and assume its values to be average, but that may lead to very uncertain and unreliable results (or not) and generally your power to see real change in your data will be poor. I am not entirely sure how diffSplice handles groups with just one replicate (i guess you can read the documentation or ask on a voom fourm).

But bottom line, i don't reccommend doing an analysis with just one replicate but it isn't strictly necessarily incorrect.

Hope that helps