LDpred2: strategy for per-chromosome resolved PLINK .bed files?

[espenhgn]

What should be our strategy for dealing with genotypes split across multiple files?
So far we've assumed a singular prefix.{bed|fam|bim} file set for LDpred2, but these can be split per chromosome.
Should we:

1. Merge using PLINK prior to predictions?
2. Extend createBackingFile.R script to allow for a list of files producing a single .bk/.rds file set?
3. Treat files separately, compute scores per chromosome, and sum the predictions. LDpred2 author implies this would be ok (Combining chromosomes from .bed files privefl/paper-ldpred2#4 (comment))

The final option would allow for trivial parallelization.

[deepchocolate]

I think 3) is a bit messy as there would have to be 2 files for each chromosome and we would have to rewrite the PGS script. Feels like it will increase complexity.

I think I'd vote for 2. Maybe we could allow for an @ parameter in the flag for the bed file to the createBackingFile.R script. Another flag like --merge could then tell the script to put all genotype data in the .bk/.rds files.

The only drawback with 1 I can think of is that it would add a plink-step whose only purpose is to make the creatingBackingFile.R-script work as intended.

[espenhgn]

Actually, for option 3 we won't need to modify anything in the R scripts; but add a Slurm job-array script template that distributes the tasks per chromosome (run createBackingFile.R and ldpred2.R per chr independently) plus another simple script that reads in the per-chr predictions and then sum the contributions.

[github-actions]

This issue appears to be stale due to non-activity