RNA-Seq, lower coverage shows more differential expression

[bob-loblaw]

But why do this? All well designed software for differential expression analysis will account for varying library depth.

Thats a good point, although the poster didn't say how they were comparing the libraries.

[kmcarr]

I have a question somehow related to this discussion.
If I have libraries to compare that have been sequenced with different deep:
example:
Lib1; 12,000,000 reads
Lib2; 17,000,000 reads
Lib3; 9,000,000 reads

It is ok to randomly pick up 9,000,000 reads from the Lib1 and 2 raw data?

But why do this? All well designed software for differential expression analysis will account for varying library depth.

[bob-loblaw]

I have a question somehow related to this discussion.
If I have libraries to compare that have been sequenced with different deep:
example:
Lib1; 12,000,000 reads
Lib2; 17,000,000 reads
Lib3; 9,000,000 reads

It is ok to randomly pick up 9,000,000 reads from the Lib1 and 2 raw data?

I think so, assuming that aside from varying coverages they are all prepared in exactly the same way.

[colaneri]

Normalizing libraries sequenced a different deep

I have a question somehow related to this discussion.
If I have libraries to compare that have been sequenced with different deep:
example:
Lib1; 12,000,000 reads
Lib2; 17,000,000 reads
Lib3; 9,000,000 reads

It is ok to randomly pick up 9,000,000 reads from the Lib1 and 2 raw data?

[bob-loblaw]

Hi Wolfgang,

I've located at one problem in my workflow that resulted in only fraction of my reads mapping. I'm hoping that this problem can be attributed to this. I'm currently re-processing my samples now. Thanks though!

[Wolfgang Huber]

Dear bob loblaw

thank you. The behaviour you report is not reasonable, and somewhere in your workflow or tool chain there must be a mistake. Can you report the sequence of steps (script) that you perform, to reproduce your observation?

Best wishes
Wolfgang

[bob-loblaw]

Hi All,

I've recently being assessing some RNA-Seq data that I have from a pilot project. When we begin our actual study though we'll have much more samples and hence we want to use less coverage to save money where possible. Basically we're looking at differential expression between 2 conditions, however when I randomly extracted 1/3 of the reads for each file and mapped them (using the exact same pipeline as the whole files) and then looked at differential expression, I found about 4 times more differentially expressed genes than with the all of the data.

Any ideas why? I've been doing this using DESeq

I also performed some clustering and found that the samples from the pilot study tend to fall into 2 fairly distinct groups, but look at differential expression within those groups isn't viable because of the sample size (only 3 samples in each group) and so DESeq doesn't detect anything as being significantly differentially expressed.