Merging RNA-seq sample with several replicate in which step?

Question

[star]

I have some samples with 3 technical replicates, I have aligned each one and also got count table using HTSeq for each one. Now, I would like to know, it is better to merge 3 replicates together before getting count table (after alignment and then get count table using HTSeq) or after getting count table for each one using HTSeq?

I think merging is not proper, and I think one of below options could work:

• select better sequence from different replicates of same sample based on number of ambiguous counts or Qc of alignment?

• Merging and getting average?

Thanks a lot in advance!

[GenoMax]

some samples with 3 technical replicates

Technical replicates have no value for HT sequencing data. Your count tables for three must be mostly similar (unless you have large variations in read numbers per sample).

Do you rather have three biological replicates, which would be very valuable?

[star]

my count tables for technical replicates are different. Also I have 3 biological replicate that I should merge them (they also should be merged after count table or after alignment?)

[h.mon]

Also I have 3 biological replicate that I should merge them

No, you should not merge biological replicates.

[star]

Thank you. but how about technical replicate, what is your opinion?

Answer 1

[swbarnes2]

It turns out that at least for Illumina samples, technical replicates aren't really needed. The Illumina prep doesn't add much in the way of technical variance.

DESeq2 has a method for collapsing technical replicates, in which case, you would do the counts on each one separately, then let DESeq merge them. But I think you'd be okay merging first, then counting. No need to average, just add all the counts together.

Never merge biological replicates. Those must remain separate to know the biological variance.

[h.mon]

It sounds like OP did technical replicates for some samples and not others. For starters, this situation should be avoided.

Now that he has technical replicates, I would process then separately, check if they are really consistent with each other (PCA plot should be fine for that), and merge the counts only after checking no batch effects are found.

[star]

@ swbarnes2 :Thanks a lot. So you mean is that merge technical replicate together after alignment and then get count table?