Splitting large imputation target cohort into smaller cohorts

[gtdoctor]

Hi again,
Imputing a large cohort, I am facing problems associated with the availability of low-priority instance workers.
Could you confirm that with SHAPEIT and IMPUTE5, neither the phasing nor imputation of an individual sample is affected by the other samples in the target cohort? - So that phasing and imputing samples N1..N10,000 as ten batches would give identical results compared with doing so in one batch ? (Apart from some of the VCF meta-data which e.g. INFO score, which can be recalculated once the samples are merged)? Or is it more complicated than this??
Best wishes,
Gabriel

[srubinacci]

Hi,
Yes, we also noticed a increase of usage on the RAP, and it's indeed harder to get low-priority instances.
Imputation is only using information from a reference panel, therefore, you can split individuals into chunks.
Phasing is a bit different, and standard reference-based phasing usually uses all individuals (in the target and reference set).
Hope this helps,

Simone

[gtdoctor]

Thanks, i did wonder about the phasing step so glad that I checked.
I will share with you my own post-processing which recalculates the MAFs and INFO scores on chunks prior to concatenation.