Differnt results with old and new DESeq package for RNAseq

[biofreak]

hi
I had done some RNAseq analysis using the older (1.5) version of DESeq and obtained differentially expressed genes. I re-run the analysis on the same data using version 1.8.2.
The results are completely different. Nothing (# of genes, Pvalues) is matching with the previous results! Moreover, it is mostly a different "set" of genes. I can not trust the results if they are not reproducible.
Can anyone please help?

[chris]

Yup. Been there... I had some genes going from p=1e-74 to p=0.01. I suggest you check the significant genes and see if they look believable in relation to the raw data. Particularly watch out for genes which are highly variable between replicates.

It's not unique to DESeq either. Others have reported significant differences in results when using different versions of the same package.

Which goes to show that reporting version numbers of tools used for analysis *must* be stored and reported.

[biofreak]

thank you for the prompt reply. I'll dig more into raw data like you suggested.
But what bothers me is that there is no correlation with the previous data! At least in my case. All the analysis following this ( functional enrichment , clustering) is going to change and so are your conclusions... It is difficult to trust the results!

[Simon Anders]

Please see these posts for an explanation:

http://seqanswers.com/forums/showthread.php?t=11839

https://stat.ethz.ch/pipermail/bioconductor/2011-December/042441.html

[biofreak]

Simon,
thanks for replying.
Yes. I actually I had asked you the similar question before and you pointed me to the development version of DEseq (then) which takes care of those high variance and small P values situations.

http://seqanswers.com/forums/showthread.php?t=13165&highlight=biofreak

So I was under impression that I am already using the modified version of DESeq until recently when we upgraded our systems, I installed the latest version of DESeq and started obtaining completely different results.

[Simon Anders]

I guess you'll have to dig and find out which version you where using when to be sure.