Is it necessary to do the same mapping/assembly for different samples?

[dawn1313]

Hi,

I am using tophat for the alignment of raw reads to a reference. I have total six samples, and half got around 70% concordant pair alignment rate whereas other half below 40% in the files of align_summary.txt.

I wonder if I need to adjust either the tophat parameters or additional QC processes to try to increase those low alignment rate samples. Because I think all the processes (including mapping and assembly) for each sample should be SAME and thus no bias for the final differential expression. Is it correct?

Thanks!!
Dawn

[bioBob]

What is the opposite of TMI?

Lots of things can lead to issues in mapping. Sample degradation, sample contamination, etc. Seems to me that one of the things I would do is to get the reads that are not aligning and check a bunch of them out, are they short, blast to something, all A or T or etc.

You mentioned assembly, is the reference a denovo transcriptome assembly? Did you use all the samples? Truly denovo or reference guided? What QC steps do you do?

[westerman]

Because I think all the processes (including mapping and assembly) for each sample should be SAME and thus no bias for the final differential expression. Is it correct?

To answer your direct question: yes, using the same processes for each sample should not cause a bias.

However going along what bioBob wrote, you have something amiss. The suggestion of taking a handful of non-mapping reads from both the 70% mapping and 40% mapping samples and see what they blast-match to should be informative. Looking harder at your QC stats would be useful.

[dawn1313]

Thanks for your reply. I didn't do QC filtering and I started with Tophat because it seems quite good and similar qualities measured by FastQc for all six samples.

I wonder if the low mapping could result from the biological cause (one gene knocked down vs wild type) and in this case I just go ahead to do assembly.

Thanks!
Dawn

[Brian Bushnell]

If you are experience such extremely low alignment and pairing rates with TopHat, I suggest you try BBMap, which is much more sensitive. Low alignment rates caused by errors and differences between the reference and the subject will incur bias even if your process is the same, because alignment sensitivity is not continuous, but stepwise.