Applying canu's trio binning function using assembly's pseudohaplotypes

[tododge]

Hi Canu team,

I’m trying to assign individual long reads to each parental (pseudo-)haplotype without parental data.

Standard approaches based on small variants (e.g., whatshap haplotag) don’t seem to work well for complex or highly unbalanced structural variation. Intuitively, a read-binning strategy using haplotype-specific unique k-mers should perform better across all variant types (SNVs + SVs).

I’m wondering whether splitHaplotype is appropriate for this use case, and if so, whether you have recommendations for parameter tuning or database construction.

So far I've tried running a couple versions of the following code. hap1.fa and hap2.fa are pseudohaplotypes generated by hifiasm for a diploid species.

At this stage, I’m not very concerned about haplotype switch errors in the assemblies.

Reads are PacBio HiFi.

meryl count k=63 hap1.fa output hap1.k63.meryl
meryl count k=63 hap2.fa output hap2.k63.meryl

meryl difference hap1.k63.meryl hap2.k63.meryl output hap1.k63.only.meryl
meryl difference hap2.k63.meryl hap1.k63.meryl output hap2.k63.only.meryl

splitHaplotype -R hifi.fq.gz \
-H hap1.k63.only.meryl 1 canu.k63.only.hap1.fq.gz \
-H hap2.k63.only.meryl 1 canu.k63.only.hap2.fq.gz \
-A canu.k63.only.unk.fq.gz

The output files (hap1 and hap2) both contain reads, but they appear to be approximately 50:50 mixtures of both haplotypes.

Any guidance would be greatly appreciated. Thank you!

[skoren]

I'd suggest trying merqury to build and QC the databases you have from the pseudohaplotypes. My concern is since they're both mixes of both haplotypes then the markers might be sparse and not very informative. Since you have assemblies you could partition the reads based on alignment as well, the marker approach will be less sensitive than mapping by definition but is necessary when you don't have an assembly.

My main question would be what the goal of partitioning the reads in this way is?