rsomics-call

rsomics-call is the alignment-likelihood and lightweight small-variant calling product in the rsomics portfolio.

The implemented product surface contains three commands:

rsomics-call pileup
rsomics-call call
rsomics-call run

pileup writes likelihood VCF/BCF from one or more coordinate-sorted alignments, call consumes likelihood VCF/BCF, and run executes the same typed stages without serializing an intermediate file. For example:

rsomics-call pileup -f reference.fa alignments.bam -Ou -o likelihoods.bcf
rsomics-call call likelihoods.bcf --model multiallelic -v -Ob -o calls.bcf
rsomics-call run -f reference.fa alignments.bam -v -Ob -o calls.bcf

All three commands use the shared rsomics-help command tree and rsomics-common result and output contracts. BED, VCF, and tabular region files use the shared rsomics-intervals coordinate contract. Variant data may stream to standard output, while named outputs are committed transactionally. --json requires named variant output so the machine result envelope remains a separate standard-output stream.

The current code establishes coordinate-merged SAM/BAM/CRAM input, multi-input sample projection, bounded reference caching, and streaming SNP likelihoods with per-input depth limits, deterministic deep-sample selection, and allele-aligned site and sample quality evidence. An explicit reference-free mode emits N reference alleles and rejects BAQ and indel configuration at the builder boundary. Indexed region input merges records across samples by coordinate, sorts selections by alignment-header order, merges overlapping or adjacent intervals, and emits each selected site once. Streaming inclusion targets require no alignment index, normalize overlapping selections, preserve alignment order, and intersect with indexed regions when both are present. Target files accept BED, one-based tabular, and VCF coordinates with transparent plain, gzip, or BGZF input. The library includes full and partial BAQ, insertion and deletion likelihoods, sample-specific reference consensus, glocal realignment, STR penalties, pooled or per-sample candidate support, explicit ambiguous-read allele-depth policy, typed indel annotations, zero-copy, haploid, diploid, and independently grouped multiallelic calls, a fused typed calling path, and strict likelihood and called-record streams for plain VCF, BGZF VCF, raw BCF, and BGZF BCF. The command boundary also binds alignment lists, sample projection, indexed region files, streaming target files, flag filters, overlap and depth policy, BAQ, indel policy, caller models, ploidy, grouped samples, gVCF blocks, prior-frequency tags, and all four output encodings. Pileup records carry the bcftools default bias metrics plus strand-aware allele depths, quality means and sums, strand bias, mismatch, and soft-clip annotations at sample and site scope. These paths are checked against bcftools and HTSlib 1.24. Called records retain those annotations, replace the internal I16 evidence with model-specific DP4, MQ, and PV4, and preserve allele-indexed fields when callers trim alternates. Typed ploidy resolution supports constant, GRCh37, GRCh38, and checked custom definitions with sex-aware or fixed sample assignments and allocation-free repeated queries. It also groups consecutive reference calls into bcftools-compatible gVCF depth blocks with typed END and MIN_DP output. Likelihood sample selection is projected during record decoding, while call sample files preserve requested order and bind explicit sex or fixed ploidy through the same definition. Multiallelic workflows can select alleles independently for validated sample groups, retain unused alternate dimensions at variant sites, and incorporate validated integer panel allele counts from configurable prior-frequency INFO tags. Prior counts remain aligned when output alleles are projected. The call stream applies masked-reference, variant-only, and SNP/indel skip policy before serialization with bcftools-compatible record-type semantics. Call sample files currently accept exactly one sample name per row; the contradictory bcftools 1.24 optional sex and numeric-ploidy behavior is not exposed as a command-line promise. Call-stage targets, target complements, and a separate unseen-allele switch are likewise absent until their documented and installed bcftools 1.24 behaviors can be reconciled. Those isolated options remain explicitly excluded from the implemented commands. The release benchmark and reproduction procedure are recorded in PERFORMANCE.md.

The historical rsomics-vcf-mpileup and rsomics-vcf-call repositories are implementation and fixture sources, not dependencies. Their single-sample shells are not the product boundary.

The multiallelic allele-selection, genotype, and quality model follows bcftools 1.24 mcall.c, Copyright Genome Research Ltd., under its MIT/Expat license retained in LICENSES/BCFTOOLS-MIT.txt.

The consensus allele-frequency posterior follows bcftools 1.24 ccall.c and prob1.c; its attribution and license are retained in THIRD_PARTY_LICENSES.md.

The established indel likelihood path follows bcftools 1.24 bam2bcf_indel.c and str_finder.c, and HTSlib 1.24 probaln.c; their attribution and licenses are retained in THIRD_PARTY_LICENSES.md, LICENSES/BCFTOOLS-MIT.txt, and LICENSES/HTSLIB-MIT.txt.

Pileup annotations and caller-side bias tests follow bcftools 1.24 bam2bcf.c, ccall.c, and mcall.c; Fisher exact and related numerical kernels follow HTSlib 1.24 kfunc.c. Their MIT notices are retained in the same license files.

Ploidy presets, custom definitions, and sample binding follow bcftools 1.24 ploidy.c and vcfcall.c under the retained bcftools MIT notice.

gVCF reference blocking follows bcftools 1.24 gvcf.c under the same retained bcftools MIT notice.

The revised MAQ error model follows HTSlib 1.24 errmod.c, Copyright Broad Institute and Genome Research Ltd., under the MIT/Expat license retained in LICENSES/HTSLIB-MIT.txt. Its deterministic drand48 sequence follows the FreeBSD implementation carried by HTSlib; the notice is retained in LICENSES/FREEBSD-RAND.txt. This product is licensed under MIT OR Apache-2.0.