Calling variants when mapping reads from different species than reference

[gruberjd]

Hi all,

My SNP calling pipeline seems to be poorly performing when tasked with mapping reads from a sample that is from a related species from the reference assembly. (This sort of thing is common in plant breeding.) Obviously the more variants that differentiate reference and sample, the harder it is to map reads correctly, so it makes sense that this is harder than finding intraspecies SNPs.

Nonetheless, I am 100% confident it can do better than it is doing. I can see from the bam that there are reads mapping to a given region, and even then most of the time my SNP caller (the last free version of GATK) fails to emit genotypes, even of sites that simply match the reference. Interestingly, the few times that it does emit a genotype, it is highly highly biased towards A&T calls.

Is any of this a classic symptom of a parameter I need to adjust, or a sign that I need to switch SNP callers? Can anyone cite literature that compares SNP callers' performance (or influence of different parameters) in this particular challenging task? I haven't had much luck searching.

Thanks!

Jonathan

[Brian Bushnell]

I don't have any papers, but I'd be interested in seeing if you get better results using BBMap for mapping and BBMap's variant caller for variant-calling. BBMap can align at short reads at quite low identity, so it's good for cross-species alignment (you can increase sensitivity with the "minid" flag). Assuming you have paired interleaved reads in a single file, and starting with the raw reads, the commands would be something like:

bbduk.sh in=reads.fq out=clean.fq ref=adapters.fa ktrim=r k=23 mink=11 hdist=1 tpe tbo
bbmap.sh in=clean.fq ref=ref.fa out=mapped.sam
callvariants.sh in=mapped.sam ref=ref.fa ploidy=2 out=vars.vcf

For multiple samples, the callvariants command would be:

callvariants.sh in=sample1.sam,sample2.sam,sample3.sam ref=ref.fa ploidy=2 out=vars.vcf multisample

At JGI we are using CallVariants a lot these days for various strains of things mapped to a divergent reference (largely Aspergillus, E.coli, and some plants from cross-breeding experiments). I don't really do variant-calling in production any more (just for testing and development) but the people who do at JGI switched to CallVariants because, as they told me, it performs much better than their prior pipelines (GATK, FreeBayes, and a couple others), particularly for variants with low allele fractions (using the "rarity" flag). In general I think it's best to use the defaults, though of course you need to specify ploidy correctly.

[gruberjd]

Brian-- Thanks. I'll give BBMap a shot and will certainly report success if I find it!

[Brian Bushnell]

Great - I look forward to your feedback.