Question

Workaround to using normalized counts as input for edgeR

0

Entering edit mode

sharvari gujja ▴ 40

@sharvari-gujja-6614

Last seen 22 months ago

United States

Dear Gordan,

I am writing to get your feedback on using normalized count as an input to edgeR algorithm. I understand that edgeR expects raw counts as input, and I wondered if there is a workaround in the pipeline to use normalized counts instead. Can you please advice.

Thanks Shar

edger • 1.9k views

ADD COMMENT • link updated 4.5 years ago by Gordon Smyth 52k • written 4.5 years ago by sharvari gujja ▴ 40

0

Entering edit mode

What do you mean exactly by a "normalized count"? See for example https://www.biostars.org/p/112637/

ADD REPLY • link 4.5 years ago Gordon Smyth 52k

0

Entering edit mode

Hi Gordon, Thank you for the reply. By normalized count, I mean TMM normalized matrix using CPM or an output from ComBat or removeBatchEffect with batch or covariate(s) adjustment. Appreciate your feedback. Thanks Shar

ADD REPLY • link 4.5 years ago sharvari gujja ▴ 40

0

Entering edit mode

Sorry, got posted twice..

ADD REPLY • link 4.5 years ago sharvari gujja ▴ 40

score 0 · Answer 1 · 2020-06-23

0

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 9 hours ago

WEHI, Melbourne, Australia

ADD COMMENT • link 4.5 years ago Gordon Smyth 52k

0

Entering edit mode

Thank you so much, Gordon. Going through these posts I understand that your recommendation is to use limma pipeline for TMM normalized or batch corrected data as input, and to use edgeR with raw counts as input only. Please let me know if I missed anything.

ADD REPLY • link 4.5 years ago sharvari gujja ▴ 40

0

Entering edit mode

Yes, you can use limma-trend to analyse logCPM values, but batch corrected data is another consideration. Generally speaking, batch correction should be done as part of the limma or edgeR linear models. It is not correct to do ComBat batch correction and then pass the results to limma as if they were not corrected.

It is surprising that you would have TMM normalized cpm values but not the original counts.

ADD REPLY • link 4.5 years ago Gordon Smyth 52k