clintval / chaff / 37689554814

97%

push

github

feat: take copied damage priors from a duplex BAM's single-strand profile (#4)

* feat: add lesion classes, lesion strands, and CpG context

* feat: add template geometry, read floors, and the molecule evidence seam

* feat: port the end repair fill-in and a-tailing filters with a learned prior

* feat: add the strand-aware lesion copy filter

* feat: add VCF and BCF IO, reference context, and the metrics table

* feat: merge-join calls with molecules, learn priors, and write annotations

* feat: add the filter subcommand with fgbio-compatible flags

* fix: order a-tailing before end repair fill-in as fgbio does

* test: port fgbio's FilterSomaticVcf tests against fgbio 4.1.1 values

* docs: describe the filters, priors, and differences from fgbio

* docs: drop the AI development note from the readme

* feat: stream the BAM through streampile's pileup engine

* fix: write metrics values at six significant digits

* feat!: require the MC tag for template ends and never read TLEN

* feat!: make chaff a single command without subcommands

* feat!: rename the lesion copy filter to copied damage

* fix: refuse options of filters that --filters leaves out

* test: run every cucumber scenario

* chore: describe the evidence layer and build profile plainly

* docs: show an end-repaired, a-tailed duplex library in the help

* docs: rewrite the readme around runnable examples on fgbio's test data

* test: check the readme examples against the binary

* ci: explain the coverage job's --follow-exec in one line

* fix: exit 2 on conflicting options as clap does

* fix: refuse --ref without the copied damage filter

* fix: exit 2 when --ref is missing or the input is standard input

* build: pin streampile to 5792224 for its template distances

* feat!: count distances from template ends in template bases

Every distance from a template end now comes from streampile's five_prime_distance and template_end_distance, which walk both reads' CIGARs, the mate's from MC, ... (continued)

632 of 675 new or added lines in 8 files covered. (93.63%)

2211 of 2284 relevant lines covered (96.8%)

366755.02 hits per line