Analysing RNA-Seq from Scratch?

[Alex234]

Hello, I am trying to find DE genes under two different conditions (I have three replicates for each). So far I have followed the TopHat->Cufflinks->Cuffdiff workflow, but the found some of the genes that CuffDiff said were 'significantly differentially expressed' (using mySigGeneIds<-getSig(cuff_data,alpha=0.05,level='genes') were higher in replicate 1 WT than replicate 1 KO, but lower in replicate 1 WT than replicate 2 KO (as judged by FPKM). Unless I'm missing some very basic statistical trick, this says to me that the difference between WT and KO conditions is not reproducible, despite CuffDiff deeming it significant.

After this I tried to find out how CuffDiff calculates variance, p-values, basic things etc, but couldn't find any equations. So I stopped trusting CuffDiff, and was thinking about trying to use R to calculate the variance p-value for each gene (using the FPKM values that CuffDiff provides for each replicate), but was wondering how much programming this would actually entail? I don't want to do anything fancy, just see whether there are any significantly and reproducibly (between all possible WT and KO replicate pairs) DEGs and be able to see the maths!

Would appreciate any advice or alternative programs / CuffDiff tips if there is a way to make it more transparent.

Thanks!

Alex

[mastal]

Have a look at the Bioconductor packages DESeq and edgeR,
and read the vignettes that come with each package, for examples of
code to do your analyses.

http://www.bioconductor.org/

[Alex234]

Thanks, I'll give it a go!