Replace per-sample FreeBayes + bcftools merge with joint multi-sample FreeBayes

[jbrestel]

Summary

• Replaces the per-sample FreeBayes → bcftools merge pattern with a joint multi-sample FreeBayes call across all samples, eliminating post-hoc merge artifacts
• Genome is split into fixed-size chunks (chunkSize = 1000000) via makeRegionBed; FreeBayes runs per-chunk across all sample BAMs in parallel
• Zero-coverage splits are derived from the union of all sample BAMs per region, ensuring no sample has a reference block that another cannot support
• The joint multi-sample gVCF is the single source of truth for consensus FASTAs, indels TSV, and per-sample CNV VCFs

New processes (modules/snp.nf)

• makeRegionBed — splits reference into fixed-size BED windows
• makeMultiSampleZeroCoverageBed — computes union zero-coverage BED across all sample BAMs for a region
• concatMultiSampleVcf — gathers per-region split gVCF chunks into a single joint multi-sample VCF; emits both gVCF and variants-only VCF
• extractSampleVcf — extracts per-sample VCF (+ SNPs + indels subsets) from the joint VCF for CNV downstream steps

Modified processes

• freebayesMultiSample — now accepts a regionLine input and --region flag; emits regionKey for scatter-gather joining
• splitGvcfAtZeroCoverage — removed per-sample BAM input; now takes a pre-computed zero_cov.bed from makeMultiSampleZeroCoverageBed
• makeConsensusFromGvcf — removed sampleName input; Python script iterates all samples in the joint gVCF, outputs one {sampleName}_consensus.fa.gz per sample
• makeIndelTSV — now takes joint multi-sample gVCF; loops over all samples via bcftools query -l
• makeSnpDensity (cnv.nf) — input tuple slimmed from 9 to 7 elements (no gVCF paths)

Removed processes

freebayes, mergeVcfs, makeMergedVariantIndex, mergeGvcfs, bcftoolsMpileupGvcf

Python script changes (bin/)

• splitGvcfAtZeroCoverage.py — replaced --bedgraph + DP recomputation with --zero-cov-bed (pre-computed union BED)
• makeConsensusFastaFromGvcf.py — updated for multi-sample: build_consensus takes sample_idx, main() loops over all samples with VCF reopen per sample, writes to --output-dir

Test coverage

• testing/t/test_splitGvcfAtZeroCoverage.py — 5 pytest unit tests
• testing/t/test_makeConsensusFastaFromGvcf.py — 5 pytest unit tests
• Stub run validated: all 8 new processes appear in DAG, no dead processes, no DSL errors

Test plan

• Run python -m pytest testing/t/test_splitGvcfAtZeroCoverage.py testing/t/test_makeConsensusFastaFromGvcf.py -v
• Run stub: nextflow run main.nf -entry processSingleExperiment -profile processSingleExperiment -stub
• Verify all 8 expected processes appear; none of the 5 removed processes appear
• Run against real diploid test data and confirm joint gVCF, per-sample consensus FASTAs, and indels TSV are produced correctly

🤖 Generated with Claude Code