Question

Variance filtering prior to RNA-seq differential expression

0

Entering edit mode

chris86 ▴ 420

@chris86-8408

Last seen 5.0 years ago

UCL, United Kingdom

Hi

I want to variance filter my RNA-seq data to increase statistical power similar to what I have seen done in microarray studies where the data groups are pretty similar. However I cannot use limma/deseq etc after the variance filter because it screws up the background variance, is it OK to do a t-test? I think there are problems with the normal distribution not fitting RNA-seq data, in which case, what to do? Any ideas?

Cheers.

sequencing deseq limma • 3.8k views

ADD COMMENT • link updated 9.4 years ago by Gordon Smyth 51k • written 9.4 years ago by chris86 ▴ 420

score 3 · Accepted Answer · 2015-07-20

3

Entering edit mode

Gordon Smyth 51k

@gordon-smyth

Last seen 1 hour ago

WEHI, Melbourne, Australia

Don't do variance filtering -- then the problems you mention will all disappear.

Variance filtering is hardly ever a good idea, and almost certainly not with RNA-seq data.

ADD COMMENT • link 9.4 years ago Gordon Smyth 51k

0

Entering edit mode

I' new to rna seq analysis. So pardon my ignorance.

Why is variance filtering hardly ever a good idea, and almost certainly not with RNA-seq data?

What if your downstream analysis is a clustering algorithm instead of differential expression?

Best

ADD REPLY • link 9.2 years ago nickbern92 • 0

0

Entering edit mode

For clustering that is fine, it makes sense. For differential expression many of the packages do variance sharing to increase power so you cannot remove the low variance ones. Actually I compared variance filtering + t test with limma for finding DE genes, and limma came out on top because it does this information sharing.