[jiwu2573]

Another question is about samples size.

I have 2 groups with case and control (paired or unpaired). Samples in each group are from different individuals. I get about 10-13 million reads/sample with 76 bp read length.

How many individuals do I need for each group to get reliable statistics? Such as power and significant DE genes.

Do I better off have balanced group (case number = control number) or I can have, say 13 cases and 7 controls instead?

[Xi Wang]

Hi Xi,

The samWrapper also works well. But I am a bit confused about the column designation.

When I looked at the output "gene.exp", 1st col gives the gene ID, 2nd col for raw counts, 3rd col for RPKM, and 4th for all reads, then the pattern keeps on repeating for each sample. In this order, expCol1 should be col 2 and 5 (raw counts columns), but how comes 2 and 4 as you wrote?

Also I don't understand measure1 or measure2. What's the use of them? By the way, my samples are unpaired.

Sorry, I have not tested the code. I think you understand the usage of samWrapper, go ahead to use it.

I am sorry again I did not notice your samples are not paired. So you may use

measure1=rep(1, length(expCol1)),
measure2=rep(2, length(expCol2)),
paired=FALSE

It's also the default.

[Xi Wang]

Another question is about samples size.

I have 2 groups with case and control (paired or unpaired). Samples in each group are from different individuals. I get about 10-13 million reads/sample with 76 bp read length.

How many individuals do I need for each group to get reliable statistics? Such as power and significant DE genes.

Do I better off have balanced group (case number = control number) or I can have, say 13 cases and 7 controls instead?

If you have few samples, the more the better. And I also think the balanced groups and even paired samples are much better.

[jiwu2573]

Hi Xi,

Can you suggest a minimum sample size that is normally acceptable for DEGSeq?

[Xi Wang]

Hi Xi,

Can you suggest a minimum sample size that is normally acceptable for DEGSeq?

I thinks 5 vs 5 case-control samples are ok. There is a tradeoff between the cost and the statistical power to convince others.