Add additional_samples_to_drop option to run_pca_with_relateds

[klaricch]

-Resolves issue where the autosomes_only filter was ignored if related_samples_to_drop was provided to run_pca_with_relateds
-Adds an additional_samples_to_drop option to run_pca_with_relateds, which will remove additional provided samples when generating the PCs and then project them in the PC space

[jkgoodrich]

Mostly one comment that will change a few names and comments/docstrings if you agree.

[jkgoodrich]

One thought is to modify this to additional samples to drop or something like that. Right now for removing unreleasable samples we just combine them with the related sample HT that we pass to this function, but we could just pass that table here too.

[klaricch]

I like additional samples to drop. Don't we drop the unreleasable samples and non-high-quality samples completely before running this step?

[jkgoodrich]

Nope https://github.com/broadinstitute/gnomad_qc/blob/a828b1529a439fc41ecc73647a442c7e63897701/gnomad_qc/v3/sample_qc/sample_qc.py#L457

    if not include_unreleasable_samples:
        logger.info("Excluding unreleasable samples for PCA.")
        samples_to_drop = samples_to_drop.union(
            qc_mt.filter_cols(
                ~project_meta.ht()[qc_mt.col_key].releasable
                | project_meta.ht()[qc_mt.col_key].exclude
            )
            .cols()
            .select()
        )
    else:
        logger.info("Including unreleasable samples for PCA")

    return run_pca_with_relateds(qc_mt, samples_to_drop, n_pcs=n_pcs)

We do drop samples that fail hard filtering

[klaricch]

ah ok, for the subpop analysis the unreleasable samples are dropped

[jkgoodrich]

Yeah, I think that's fine

[jkgoodrich]

looks good to me, only thing is that we might want to add a breaking change label because if anyone used run_pca_with_relateds(qc_mt, related_samples_to_drop_table, n_pcs) it won't work