One cool thing epi2me/wf-human-variation does is make chunks based on sample depth: https://github.com/epi2me-labs/wf-human-variation/blob/bb8b526d97e0bbb169ba0fa3596eb5d844b927a8/modules/local/wf-human-snp.nf#L51-L69 So you align the BAM, then create chunks ONLY where there is sufficient depth. Could be useful for making Sarek more efficient.

Reminds me of the “callable regions” bcbio calculated. We’re doing something similar in our nf-cmgg/pipeline where we take the depths from mosdepth and intersect with the WES ROI

Oooh

@Lorena Pantano will be able to shed more light on the inner workings of bcbio, as she helped develop it together with Brad Chapman

Cf the plan: https://nfcore.slack.com/archives/C05V9FRJYMV/p1776086921922369?thread_ts=1776086921.922369&cid=C05V9FRJYMV

yes, nothing super fancy, I don'r remember the cutoff for coverage, but we can retrieve that from the code. bcbio would parallelize by all the regions.

Worth noting that that wf-human-variation step is actually

Clair3

(the SNP caller).

Clair3

supplies a single entrypoint which runs all these steps itself but for some reason ONT decided to implement each step directly in nextflow themselves (not using the wrapper command). There is supposedly some performance benefits but I’ve never been clear where they come from.

You could hack together a version with samtools + awk to show if it was useful