I don't know if I'm in the right place but I have a technical problem to fix. I would have to align paired-end reads from Illumina sequencing to compare a normal genome with a tumor one.

When I align with bwa-mem (default parameters) with paired-end mode I get a bigWig file where the coverage of the two samples is quite similar. When I align with bwa-mem (default parameters) with single-end mode I get coverage holes (definitely does not align any read) in some specific regions. But this only happens in tumor genomes.

Is it possible that something is really going on in that region? Or is it just a bias? And if it's a bias, why in the normal condition, in that specific region, do they align?

I think I understand that by treating paired-end reads as single-end I lose the directionality of mate2 if I align mate 1. But I still can't understand why I have this alignment defect only in the tumor genome and not in the normal genome (I observe this phenomenon in all the samples analyzed, not just in one).

Could someone give me some explanation or recommend any checks to do?

spoiler - these holes fall exactly in highly repetitive regions


2 Answers 2


In principle one should simply not align paired-end reads as if they were single end reads, not only because you are not taking advantage of the forward/reverse information contained in the reads, but also you are not taking into account the expected distance between the two pairs, both of which result in better mapping rates/accuracy.

Having said that, and not recalling from the top of my head right now the relevant BWA-mem algorithmic implementation details, it is generally difficult to map (even) paired-reads to repetitive regions. Still there will be some rate of success. In tumor samples however, we could reasonably imagine that chromosomal rearrangements, INDELs, CNVs, etc. will make it even harder to confidently assign these reads to a location in the reference, since these extra variation will increase e.g. multiple mapping locations and mapping qualities for some fraction of the reads, thus resulting in regions with low or no coverage compared to non-tumor samples.

Repetitive regions might exacerbate the problem and get affected even further compared to other regions.


I tried to give me this explanation: with single-end alignment of paired-end reads, losing the directionality of the reads and also the expected fragment length, the R2 no longer maps if it maps the R1, but can map as a single reads anywhere, or don't map at all. I thought that therefore I don't see alignment (in tumors) in repetitive regions because of the high mutation rate I have in the tumor sample. While in normal I don't have this situation. However, aligning with paired-end reads, I see coverage in these regions again despite the high mutation rate because R1 and R2 help each other map in the best position, and so even if I have mutations I still see the alignment. Could it be appropriate to make a variant call despite being on repetitive regions?

  • $\begingroup$ As it’s currently written, your answer is unclear. Please edit to add additional details that will help others understand how this addresses the question asked. You can find more information on how to write good answers in the help center. $\endgroup$
    – Community Bot
    Commented Oct 8, 2022 at 15:02
  • $\begingroup$ Check on the topic of 'mapability' (e.g. 1000 genomes or UCSC tracks, snpable, ..). Often it is practice to mask out regions of poor mapability in order to decrease the rate of false positives. Repetitive regions fall often within poorly mappable regions, your read length becomes somewhat relevant here $\endgroup$
    – JRodrigoF
    Commented Oct 9, 2022 at 9:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.