search

amplab / snap

Scalable Nucleotide Alignment Program -- a fast and accurate read aligner for high-throughput sequencing data
https://www.microsoft.com/en-us/research/project/snap/
Apache License 2.0
287 stars 66 forks source link

better handling of complex indel situations #143

Open bwlang opened 2 years ago

bwlang commented 2 years ago

SARS-CoV-2 omicron commonly has a 3 base deletion followed by 8 bp of normal sequence then a 9 bp insertion in the S gene. This is a tough, but relevant alignment situation. Lots of omicron consensus sequences are wrong as a result of bwa-mem soft-clipping if this region is near the end of a read. Snap is pretty similar to bwa-mem... aligning a few more reads , but still missing most of the insertions. snap: 259 ins / 671 total bwa-mem: 242 ins / 638 total

About 2/3 of those look "practical" to align to me. Maybe it would be good to penalize alignments that cause frameshifts more than those that don't (in this case both indels are a multiple of 3 bp).

Any chance this can be improved? Maybe some setting adjustments?

image

Reads are available via SRA: https://www.ncbi.nlm.nih.gov/sra/SRR17132492

bwlang commented 2 years ago

@bolosky : I think you mentioned that you might be interested in this...
Just a quick update. I examined BA.1 GISAID sequences in this region recently. Thousands of genomes in GISAID are missing the insertion. Most of those are also missing or substituting Ns for the 3bp deletion too. I think consensus generation tools would have a much easier time if the read stacks were more consistent. I suspect that most actually do contain these structural features but it will be very hard to tell if the genome is shifting in this area with current consensus data.

kokyriakidis commented 2 years ago

@bwlang What's your general impression regarding indel representation between bwa and snap2? Do you think that snap does a better job? Have you found other cases that on mapper performs better than the other?

bwlang commented 2 years ago

@kokyriakidis : I think they are pretty similar... but the situation could be improved with an option to consider framing (i.e. 3bp deletions are less penalized than 2 or 4 base deletions).