feat: optional BAM → CRAM checkpoint after MarkDup / BQSR

[abhi18av]

Summary

Pattern 1 of the BAM-checkpoints spec (abc-universe/specs/active/magma-bam-checkpoints.md) — optional BAM → CRAM checkpoint at the post-MarkDup-or-BQSR junction, opt-in via param, default off. Genozip / zstd variants will land as separate PRs once this has run on a real cohort.

Architecture — drop-in module swap with identical channel shape

call_wf.nf currently runs SAMTOOLS_INDEX at the post-MarkDup-or-BQSR junction, emitting (sampleName, *.bai, *.bam) which HC and the minor-variants HC consume directly. With --bam_checkpoint_compression cram, a new SAMTOOLS_VIEW_CRAM module runs instead and emits the identical channel shape (sampleName, *.cram.crai, *.cram). Downstream consumers are agnostic — GATK HaplotypeCaller, the minor-variants HC, and LoFreq (htslib-backed) all read CRAM directly via -R / -f <ref>, zero round-trip cost at consume time.

def aligned_indexed_ch
if (params.bam_checkpoint_compression == 'cram') {
    SAMTOOLS_VIEW_CRAM(recalibrated_bam_ch, params.ref_fasta, [params.ref_fasta_fai, params.ref_fasta_dict])
    aligned_indexed_ch = SAMTOOLS_VIEW_CRAM.out
} else {
    SAMTOOLS_INDEX(recalibrated_bam_ch)
    aligned_indexed_ch = SAMTOOLS_INDEX.out
}
// ...HC, minor-variants HC, LoFreq, samtools_bam_ch emit all read aligned_indexed_ch

One if/else, no new emit shape, no changes to any of the 4 processes that read the channel.

Files

Path	Purpose
modules/local/samtools/view_cram.nf	NEW. samtools view -C -T <ref> --output-fmt cram,version=3.1 + samtools index. Lossless by default; opt-in lossy quality binning via params.cram_lossy_qualities. Same samtools binary MAGMA already uses.
workflows/call_wf.nf	Branches on params.bam_checkpoint_compression at the post-MarkDup-or-BQSR junction. Routes the resulting aligned_indexed_ch through HC + minor-variants HC + LoFreq + cohort emit.
default_params.config	Adds bam_checkpoint_compression = 'none' (default) and cram_lossy_qualities = false.
CHANGELOG.md	unreleased entry with the flag matrix and rationale.

Defaults

bam_checkpoint_compression = 'none' keeps SAMTOOLS_INDEX in the pipeline; the new module isn't even loaded into the DAG. Existing invocations produce identical outputs (plain .bam + .bai) to master. Opting in is a single flag.

Pre-merge gates

• ✅ nextflow lint modules/local/samtools/view_cram.nf — clean
• ⚠️ nextflow lint workflows/call_wf.nf shows a pre-existing addParams() DSL2-deprecation warning at line 26 (the first include, untouched by this PR) — not introduced here

Open items (documented in spec, not in this PR)

• Round-trip equivalence test — compress → consume → compare HC output against a non-CRAM baseline run on the same cohort. Spec §8 conformance gate; not blocking this PR's opt-in default but blocking any "switch default to cram" follow-up.
• SV path (structural_variants_analysis_wf.nf) uses its own MarkDup / recal chain and is not touched by this PR. CRAM extension to DELLY's BAM input is a follow-up if needed.
• Samplesheet routing for pre-compressed inputs (.cram on input) is Pattern 4 of the spec.
• genozip + zstd codecs — separate PRs; genozip license-gated; zstd license-free but modest gain on already-bgzipped BAM.

Refs

• Spec: abc-universe/specs/active/magma-bam-checkpoints.md
• Sibling — XBS gVCF checkpoints: abc-universe/specs/active/xbs-variant-calling-gatk-optimizations.md §3
• htslib CRAM 3.x spec: CRAMv3.pdf