Latest DESeq2 version (2.11.40.8+galaxy0) gives different result than earlier versions

, ,
1

I have observed that the latest version of DESeq2 (2.11.40.8+galaxy0) produces results considerably different compared with earlier versions (such as 2.11.40.7+galaxy0 to 2 and 2.11.40.6+galaxy1 to 2) from the same counts files, without changing advanced options. For example, in our case the number of upregulated genes with Padj <0.05 and Log2(FC) >1 were 268 in the latest version (2.11.40.8+galaxy0) compared with 85 with five earlier versions tested. It would be great if there was some clarification why this happens.

DESeq version comparisons
DESeq version comparisons1920×1440 105 KB

2

Welcome @jmt

This might be too technical, but the Galaxy wrapper was updated to use a new version of the Bioconductor release of DESeq2.

In short, the algorithm changed. In order to get “the same” results as before there are some options that need to be adjusted, and it still may not be an exact replication.

Is this enough or would you like to discuss this more with one of the scientists who helped to tune the default parameters in the Galaxy update? There might be more discussion at the Bioconductor forum, too, but I’m not sure what parts you are most interested in or if you have seen them already. Maybe try a search there about DESeq2 version 1.38?

Hope this helps and we can follow up! :scientist:

3

Thanks very much Jenna, very helpful! The discussion fits exactly what I have observed i.e. almost identical number of DE but a major change in FC. The philosophy of this change in algorithm is beyond my understanding but happy to see that it is intentional and I have not screwed up something.

1 Like
4

No worries, I am really glad that helped! Sometimes is hard to know where an observed change originated. Feel free to reach out here anytime you have a question. And if there is an actual bug, we want to learn about the corner case, investigate, and squash it!