I'm stuck on this question. I have read that I should disable the spliced alignment for Hisat2 when aligning DNA reads, but confused on the results.
So I have my illumina shotgun reads, adapters removed and host removed. I first map with Bowtie2 with Very Sensitive end to end against a reference, which leaves me mapped and unmapped. Then I map the reads again using Hisat2, as it tends to find reads that Bowtie2 misses for some reason. So since what I have read, that I should disable the spliced alignment with Hisat2 when mapping DNA reads, so I did this which again gave me mapped and unmapped reads. Then as a test on the unmapped reads I ran Hisat2 again without disabling the spliced reads option, which again gave me a good number of mapped reads. I took a handfull of the mapped reads, both forward and reverse and ran them on the Online NCBI Blast and it looked like those reads should have been mapped by either Bowtie2 or Hisat2 in the first run. I did not look at the insert size on the reads, so it could have a much larger insert size.
What is the recommendation here??? Should I not disable the spliced alignment or not??
Disable it - a spliced alignment on genomic DNA is a gap across a junction that isn't there, so those are false by construction.
I don't think splicing is what's actually happening in your third pass though. Bowtie2 --very-sensitive is end-to-end, so the whole read has to align; HISAT2 soft-clips by default. The reads HISAT2 picks up are mostly ones where part of the read doesn't match - leftover adapter, a chimeric read, something crossing a structural difference - and they'd map under bowtie2 --local too. Easy enough to test: rerun bowtie2 in local mode on your unmapped set and see how much of the gap closes.
The chaining is probably the bigger issue anyway. Reads that only map on the third aligner are the ones most likely to be junk, and you have no consistent MAPQ scale across the three tools. For DNA shotgun I'd just use bwa-mem, which soft-clips natively and would likely catch most of these in a single pass.
Probably a better solution is to use bowtie2 with --very-senstive-local settings,
that mode will soft clip the same manner as bwa does.
In general we should not fiddle with disabling a core functionality of an aligner (in this case hisat2) just to make it work like another tool.
Fair, --very-sensitive-local is the cleaner call given they've already got bowtie2 set up. And agreed on the general point - turning a core feature off so one aligner imitates another is the wrong instinct to encourage.
Log in to answer this question.