Independent filtering by using fold change in DESeq2

[icanwinwyz]

Hi everyone,
Thanks for reading my post.

I know DESeq2 uses normalized mean count as independent filtering to improve the performance of multiple testing error correction. I would like to know is it possible I can add the criteria, |log2FC| > 1, together with normalized mean count as the independent filtering? If so, how could I add this criteria? by using "filter=" or "filterFun=" argument? How to write the command?

What I know now is to change the "lfcThreshold=" to 1, but this only change the null hypothesis from 0 to 1, not actually filter any genes.

Many Thanks in advance!

[dpryan]

No you can't use fold-change in independent filtering, it wouldn't then be independent. You can filter genes after the fact for the fold-change threshold you want.

[icanwinwyz]

No you can't use fold-change in independent filtering, it wouldn't then be independent. You can filter genes after the fact for the fold-change threshold you want.

Hi dpryan,
Thanks for your reply. You mean I should run DESeq for DE analysis first and then filter out the genes with small fold change? If so, could I only perform multiple testing error correction for those genes remained instead of all the genes tested? Since this would give me more significant DE genes with big fold change.

Thanks.

[wdecoster]

But that is a violation of independent filtering principle, you cannot filter for something you are testing for. Obviously it is great for your p-values, but also completely invalid and should be considered p-value-hacking.

[icanwinwyz]

But that is a violation of independent filtering principle, you cannot filter for something you are testing for. Obviously it is great for your p-values, but also completely invalid and should be considered p-value-hacking.

Hi wdecoster,
Thanks for you reply.
So nothing we can do by using fold change to filter the genes? any suggestion? Thanks

[dpryan]

You can filter genes by fold-change after calculating p-values and adjusting for multiple testing. If you do anything like that beforehand odds are good that you'll end up with me as a reviewer and I will guarantee that your paper is rejected.