Biological replicates with very different number of reads

[mblue]

Hi,
I want to do a differential analysis between two conditions and I have 3 biological replicates per condition. All samples have similar number of mappable reads (around 5 million) except one with nearly 100 million.
My samples were multiplexed and run 3 per lane. The 100 million read sample was run alone.

Can anyone tell me if I can include this sample in my study or if I have to repeat it? What is the best approach for biological replicates where 1 is very different from the others in terms of depth?

I have been doing differential analysis of 2 conditions by adding the number of reads of biological replicates per gene and then using a binomial test to determine p values (and these to calculate FDR).
Does anyone know if I could still do this with the "abnormal" biological replicate?

Thanks a lot! Any help is very appreciated.

[schmima]

hm - 5 and 100 mio is quite a big difference. In "normal" cases, one can take different library sizes. Assuming that the larger sample is a simple upscaling of the smaller one (looks like a shifted cloud in the scatter plots), global normalisations should deal with the problem relatively well. But if this applies to such a big difference... ?

ad replicates:
If you have 3 replicates, treat them as such. Summing up the reads and comparing these sums with a binomial test is not a really smart thing. Check out some of the R-packages for differential expression (edgeR, DESeq) to get a better idea how to compare the two conditions.

[Simon Anders]

In principle, negative-binomial based tests such as that of DESeq and edgeR should be able to deal even with vastly different library sizes. Of course, the huge sample will not add much power because you have so many reads only on one side of the comparison.

If you do a binomial test, it does not matter how you deal with it because the result will be wrong anyway. (See the numerous earlier threads on why Poisson-based tests, and that includes the binomial test, are inadmissible because they ignore biological variability.)

[mblue]

preference for DESeq and edgeR?

Thanks for such straightforward answers, they were very helpful.

I will definitely try DESeq or edgeR. Any preference for one?
I am looking into publications about these methods. If you know of any good publications please share.

I'm also looking at previous threads on poisson-based methods.