Highly heterozygous reads mapping

Question

I have short (67bp) Hi-C reads from a highly heterozygous organism (~15% between-haplotype SNP divergence) and I have both reference haplotypes.

I wanted to try and compare different haplotyping softwares for Hi-C reads using these reads as benchmarking dataset. When mapping the reads separately to each haplotype, I get good mapping statistics. When I map the reads to a single haplotype with all the heterozygous SNPs masked (into N's), I get very poor mapping rates.

I would like to be able to map the reads when the real haplotypes are unknown (reference is a mix of haplotypes).

I use minimap2 to map the reads with the sr preset. I tried decreasing the mismatch penalty (-B) to 1 and increasing the --score-N value, but this had no effect.

As shown in the attached IGV screenshot, coverage drops to 0 when the SNP density increases. Is it feasible to map reads with such a high heterozygosity on a single (masked) reference ? Should I use another tool ?

Answer 1

I believe you may be able to map your reads, but I don't know how to do that with minimap2.

I recommend running gsnap, which is more SNP tolerant and provides a number of parameters which are likely to help.

For instance, I believe that most aligners will treat 'N' characters as mismatches when you align. GSNAP has parameters to account for this.

--query-unk-mismatch=INT       Whether to count unknown (N) characters in the query as a mismatch
                               (0=no (default), 1=yes)
--genome-unk-mismatch=INT      Whether to count unknown (N) characters in the genome as a mismatch
                               (0=no, 1=yes (default))

It also has a mismatch parameter similar to the one you described for minimap2.

-m, --max-mismatches=FLOAT     Maximum number of mismatches allowed

Try running with the 'genome-unk-mismatch' parameter above (with the masked reference). That may be your best bet. There are also other parameters that may help, but this should be a good start.