Question about Haplotype scaffold

[tgong1]

Hi,

Thank you for developing and maintaining this useful tool. We got a question about the haplotype scaffold can be used.

For our samples, we got a subset of variants with high-quality haplotypes, but not chromosome-length. This means this set of genotype data have multiple haplotype blocks (with PS tag) in VCF. If this set of genotype data is used as haplotype scaffold, does shapeit5 only assume chromosome-length haplotype (ignore PS tag) or it will somehow consider the existence of multiple haplotype blocks based on the PS tags ?

Thank you for the help!

[Eurphus]

Bumping this, @tgong1 did you ever figure this out?

[cmosso]

I haven't used shapeit5 with PS tagged VCF data, so I'm not speaking from first hand experience, but it might ignore the PS tag because of the updated PBWT algorithm they use for phasing common variants to create a scaffold for the remaining variants. The methods "Common variant phasing" section of their shapeit5 paper has this:
"""
However, one computational limitation of SHAPEIT4 resides in its inability to parallelize the construction of the PBWT, which can become relatively long in very large datasets. In SHAPEIT5, the main improvement we introduced for common variant phasing is a parallelization scheme for the PBWT construction: several PBWT passes are run in parallel on several central processing unit cores, each one running for a different chunk of 4 cM by default, achieving a notable reduction of the wallclock running time of the method.
"""

You might have to use some custom genome chunks and bcftools commands to achieve your goals, but I'm not 100% sure. As I understand it, a lot will depend on how much overlap the phase set variants have across your samples.

I've never used shapeit4, but the documentation for shapeit4 (https://odelaneau.github.io/shapeit4/#documentation) has a section about the PS tag and the "--use-PS" option for shapeit4 if you can get that working.

@odelaneau have you already addressed this PS tag question for shapeit5 elsewhere?

[Eurphus]

Ahh, I hope it doesn't! Thanks for the response here.

I was hoping to be able to use WhatsHap (read-based, local phasing) as a scaffold for the data, and have the PS blocks respected whilst using a reference panel.