Strategies for highly heterozygous haplotype-resolved eukaryotic pangenomes with extreme structural variation

[GitEnricoNeko]

Hi Cactus developers,

We are building a pangenome for Mytilus galloprovincialis using multiple haplotype-resolved assemblies (currently 12 haplotypes).

Our biological goal is not only genome alignment, but especially:

• localization of presence–absence variable (PAV) genes,
• structural interpretation of accessory genes,
• graph-aware gene projection across haplotypes.

However, we are encountering substantial graph complexity and fragmentation.

Current setup:

• haplotype-resolved assemblies
• very high heterozygosity
• extensive structural variation
• high repeat content
• large gene presence/absence variation

Symptoms:

• extremely dense graph structures
• many crossing paths / graph “hairball” regions
• very large node/edge counts
• difficult interpretation of local gene neighborhoods
• very slow vg giraffe mapping

We suspect this may reflect a combination of:

• biological variation,
• fragmented alignments,
• excessive local graph branching,
• and possibly the limits of whole-genome cactus alignment for extremely polymorphic eukaryotic genomes.

We would like to ask:

1. Are there recommended strategies for highly heterozygous haplotype-aware eukaryotic pangenomes?

2. Would you recommend:

   • stronger graph filtering?
   • chromosome-by-chromosome construction?
   • pre-collapsing haplotypes?
   • different minigraph/cactus settings?
   • alternative graph normalization approaches?

We would greatly appreciate any guidance or best practices for this type of dataset.

Thanks a lot!

[glennhickey]

Yeah, MC and downstream tools like vg giraffe do break down at higher divergences. I think you get a warning after about 2%. What are the mash distances reported in your log? One thing that can help with fragmentation (but will take more memory to compute) is to use --noSplit to toggle off chromosome splitting -- so inter-chrom rearrangements won't be filtered out. Otherwise you can try Progressive Cactus and use maf/taf/hal tools for your pav analysis.

[GitEnricoNeko]

Thank you so much for the suggestion.

Yes, I did get the warning, since the average Mash distance between assemblies is around 0.035, so above the ~2% threshold you mentioned.

I also tried running the pipeline with --noSplit to reduce fragmentation caused by chromosome splitting and potential inter-chromosomal rearrangements, but unfortunately the results were still not very satisfying. This is the command I used:

bash cactus-pangenome mytilus-new mgal_pangenome.seqfile \ --outDir mytilus-new \ --outName mytilus-new \ --reference Mgalref \ --vcf --giraffe --haplo \ --gfa full filter \ --gbz full filter \ --odgi full \ --chrom-og full \ --viz full \ --draw full \ --workDir tmp/ \ --refContigs $REFCONTIGS \ --otherContig chrOther \ --permissiveContigFilter \ --noSplit

So do you think that the progressive cactus pipeline can be the strategy?

Thank you again!