Question

the output DEGs of limma analyzing RNA-seq data is just 1/3 of those by Tophat-CuffDiff

0

Entering edit mode

boyashuang • 0

@boyashuang-11654

Last seen 8.2 years ago

Thanks Dr Smyth and Dr Shi for sharing limma to analyze RNA-seq data.

The first confusion is :

Does very noisy data mean having rows with zero or very low counts?

Secondly, after performing voom normalization, why the output DEGs (adj.P.Val<=0.05) of limma analyzing RNA-seq data is just 1/3 of those by Tophat-CuffDiff (q<0.05), regardless of setting the absolute logFC>=1.

Furthermore, the mountain peak of voom: Mean−variance trend is not so charming as the expected in protocol， something like smooth downhill.

So we want to know what shall we do to normalize the DataMatrix?

Can we set logFC to 1 and adj.P.Val<=0.1, with the moderate threadshold?

Thanks again.

differential gene expression voom limma RNA-seq • 1.2k views

ADD COMMENT • link updated 8.2 years ago by Gordon Smyth 52k • written 8.2 years ago by boyashuang • 0

1

Entering edit mode

You'll want to split voom and limma into separate tags, otherwise the maintainers won't get notified.

ADD REPLY • link 8.2 years ago Aaron Lun ★ 28k

0

Entering edit mode

Having rows with low or zero counts typically means either your gene/feature is low in abundance or you haven't sequenced deep enough to detect it. You need to remove rows with low or zero counts for limma to work properly I believe, this could help. Your method or normalisation depends on your intent.

ADD REPLY • link 8.2 years ago chris86 ▴ 420

score 0 · Answer 1 · 2016-10-16

I don't entirely understand your questions, but perhaps you might read the voom documentation, or follow the case study in the limma User's Guide or follow a published workflow example:

https://f1000research.com/articles/5-1408

It certainly sounds as if you have not done the minimal filtering that you are expected to do prior to running voom.