Hi all,

As part of a project I am working on in UK Biobank (UKB), I am tasked with calculating polygenic risk scores from three existing scores (PGS003725,PGS002724,PGS005217) from the The Polygenic Score Catalog for all individuals in UKB.

To this extent, I am trying to use the pgsc_calc tool which is to my understanding is widespread used for this purpose.

However, I keep running into certain unfortunate errors / extremely long computational times, that I cant really seem to wrap my head around the way to best move forward. Unfortunately, my supervisor does not have a lot of knowledge on this topic, so I am a bit left clueless. Therefore, I have a few nooby questions I was hoping you could help me out with:

1) The initial attempts I have run seem to take exorbitantly long which I have had to abbort due to computational costs. I have ran these on mem3_ssd1_v2_x16, using the custom config below. What would be the somewhat expected runtime and costs using theses settings?

  process {
    withName: 'PGSC_CALC_FORMAT' {
        cpus = 2
        memory = { 16.GB * task.attempt }
    }
    withName: 'PGSC_CALC_LOAD' {
        cpus = 3
        memory = { 120.GB * task.attempt }
    }
    withName: 'PGSC_CALC_SCORE' {
        cpus = 2
        memory = { 120.GB * task.attempt }
    }
}

2) Up to know ive used dxfuse based methods. However, the pgsc calc website reccomends the following in case of many variants: for more complex jobs, which can contain millions of variants, consider downloading the genotype data to your cloud workstation to significantly speed up calculation. Since the files are massive, I don't really know how to approach this. Is this necessary in my case and, if so, how to best approach this?

3) Lastly, arguably the most basic question, for my approach; to what extent does it matter if I use TOPMED, GEL or UKB-imputed data?

Thanks in advance!

Kind regards,

Theo C