allelix 2.4.1

v2.4.1

on Python PyPI

Added

• Reports say where the build came from: forced with --build, the
  chromosome lengths, the known-SNP positions, the file's build line or
  VCF assembly, or the format default. The JSON report has
  input.build_source and input.build_evidence (schema version 8;
  --diff still reads older reports), the HTML report shows the source
  under the build, and the annotated VCF writes
  ##ALLELIX_Build=<ID=...,Source=...> (VCF schema 0.3.0).
• --build grch36 (and --build1 / --build2 grch36 for compare):
  runs an old GRCh36 / hg18 file that Allelix cannot identify by itself,
  with position-keyed lookups off, as for a detected GRCh36 file.
• --build for stats and extract, and --build1 / --build2 for
  compare.
• MyHeritage array files: the "human reference build 37" header sentence is
  now read as the file's build label.
• VCFs whose mitochondrial chromosome has the old hg19 length: mitochondrial
  records are skipped, with a note, because their positions do not match
  the reference sequence ClinVar uses.

Removed

• scripts/tag-release.sh, a release-tagging helper, has been removed.
  Tag pushes are still checked against the pyproject.toml version by
  the pre-push hook in .githooks/.

Fixed

• Reports on the wrong genome build: files whose build label was wrong
  or missing. Allelix looked for its build-detection SNPs only in the
  first 100,000 rows of a file. Real raw data files are sorted by
  chromosome, so on full-size files the build almost always came from the
  file's label, from the way a VCF names its chromosomes (chr1 or 1),
  or from the format's default. When that guess was wrong, the report used
  the other genome build's positions: real findings were lost, and
  findings the person does not carry appeared, including ClinVar
  "pathogenic" ones. This affected chip files whose header names the wrong
  build or none, UCSC hg19 VCFs with chr-named chromosomes, and GRCh38
  VCFs with plain chromosome names, in every release since 2.0.0.
  Allelix now decides the build before annotating, from the evidence in the
  file itself: a VCF's declared chromosome lengths, or well-known SNP
  positions across the whole file. The run stops instead of guessing when
  that evidence contradicts the file's label, or when the file carries no
  usable evidence: a VCF without chromosome lengths, without an assembly
  line and without enough of those SNPs, a chip with no build line and
  none of the SNPs (such as an X-chromosome-only export), or SNPs that sit
  at none of the supported builds' positions. The message says what was
  found and asks you to name the build with --build grch37 or --build grch38 (or --build grch36 for an old GRCh36 / hg18 file). Nothing is
  written in that case.
  If you have reports from such files, run them again with this release.
• stats, extract, export plink and compare now decide the build the
  same way as analyze. compare no longer settles on a weak guess.
• Three entries in the build-detection SNP table were wrong (one SNP on
  both modern builds, two on GRCh36). They now match dbSNP. Old GRCh36
  FTDNA files that were detected only tentatively are now detected with
  confidence.
• The note about GRCh38 VCFs without rsIDs no longer claims GRCh37 gives
  more findings. It now appears only when the rsID identity table or
  gnomAD is not installed, says which, and tells you to run
  allelix db update.
• A failed export plink no longer leaves partial .bed / .bim / .fam
  files behind.
• export plink no longer writes a PLINK set with no variants. When every
  record is a no-call or an indel, or was dropped (for example mitochondrial
  records under an hg19-length chrM), it stops and says what was skipped.
• AlphaMissense and CADD caches without a schema version stamp are no longer
  treated as current. allelix db update replaces such a cache with the
  current one, as it already did for gnomAD.
• Very old GWAS Catalog and ClinPGx caches, made before Allelix recorded a
  version inside each cache, are accepted only when the version noted in
  them is the current one. Otherwise the GWAS cache is rebuilt from the
  download kept next to it, if that file is still there; if not, GWAS
  Catalog, like ClinPGx, is skipped until you run allelix db update.
• When an upgrade changes how Allelix reads ClinPGx data, allelix analyze
  no longer rebuilds the ClinPGx cache itself, which kept the old CPIC
  allele-function data. It skips ClinPGx and asks you to run
  allelix db update, which rebuilds the cache and fetches the CPIC data
  again. analyze makes no network request for it.
• The source distribution no longer ships a partial, unrunnable copy of the
  test suite (the top-level tests/test_*.py files without their fixtures).
  It contains the package, the licence, the README and the build files.
• Building Allelix from source now declares the setuptools version it needs
  (77 or later, for the SPDX licence field). With an older setuptools the
  build failed with a project.license configuration error.
• The GRCh36 warning no longer claims gnomAD results are complete: on
  GRCh36 files ClinVar, gnomAD, AlphaMissense and CADD give no values, and
  only ClinPGx, GWAS Catalog and SNPedia (matched by rsID) run. It also
  says how GRCh36 was chosen.