Sequence in (biological) duplicate or not?

[Jasseq]

Dear all,

I am wondering: how much (methodological) advantage would it have to RNAseq in duplicate (i.e. 2 biological replicates) compared to 1 single seq?

I do not have the resources to have 3 replicates (or more). Samples will be pooled set anyhow (from 6 subjects), but can pool into 1 (=6 subjects) or 2 (=2x3 subjects) samples per condition.

Of course, statistics will be still sub-optimal when have only 2 measures per condition compared to >=3 -- but on the other hand, the reliability should be better. I am mainly interested in differential expression, btw.

Thanks!

[morgane51]

I was asking myself the same thing, but I could go to 3 replicates.
Take into account that duplicates means twice as much work but not twice as much power. However, having 2 duplicates, you introduce noise but can not handle it/correct for it.
If your duplicates agree, then you gain in power. If they don't, you cannot explain it.
Merging the samples will bring some biological noise that could make you miss some signals, but on the other hand when you treat only one sample you have no way to spot incongruency due to noise.
I don't know if I am clear enough... But I think it is better to have duplicates - but do not sacrify to much on coverage... -
Still, the best improvment is from 2 to 3, and not from one to 2 replicates...
This is really well explained in the manuals for DSeq and EdgeR. They advise to refrain from going for one replicate.
What is your design exactly?
Understand that you waste the whole money at the end if your design doesn't allow for a correct analysis...
Good luck!
Morgane

[Jasseq]

Thanks for your reply.
Specific design is: Two groups, 8 brain regions, 6 individual samples per brain region (mice).
Individual samples can be pooled per 3 (i.e. 2 pooled biological replicates per brain region, 32 pooled samples in total to sequence), or per 2 (48 pooled samples to sequence).
The fact that I want to analyse 8 brain regions makes the costs-to-number-of-samples-ratio so steep...

[HESmith]

Be aware that triplicates do not require 50% more sequence data than duplicates. Given the same total number of reads, more replicates provide more statistical power to detect differential gene expression (i.e., 3 reps @ 10M reads is better than 2 reps @ 15M). See this paper for details. So the only added expense for triplicates vs duplicates is the cost of library prep.