allelix 2.6.2

v2.6.2

on Python PyPI

Changed

• The HTML report and the terminal show plausibility only when it warns.
  The muted "○" beside a call that was not assessed, its "Not assessed" row
  in the HTML detail panel and the terminal's "○" legend line are gone; the
  "⚠" (and the terminal's "?") for an implausible or suspicious call stay,
  with their panel row. The warning is explained once, in plain words, in
  the HTML "Reading This Report" section, the terminal's legend line and
  the README: the call looks doubtful because the variant's frequency in
  gnomAD is at or under Allelix's limit for it (under it for two copies; by
  default, two copies of an allele below 0.01%, or one copy of an allele
  that gnomAD lists with no carriers), and since Allelix can only check
  variants with population counts in gnomAD, no warning is not a guarantee.
  The suspicious hover text and detail-panel row give the zero-carrier case
  as the default too. JSON reports are unchanged: plausibility and
  plausibility_reason stay on every annotation (ADR-0050).

Fixed

• db clean, db update and an analyze, methylation or
  pharmacogenomics run that refreshes a cache no longer crash with
  UnicodeEncodeError printing to the console in a POSIX ASCII locale
  (LC_ALL=C and PYTHONUTF8=0, with PYTHONIOENCODING unset; on Python
  3.12 and later LC_ALL=C alone turns on UTF-8 mode, which did not crash),
  where the console's error mode is surrogateescape, not strict. db clean used to crash after it had deleted the caches, and db update --force and the refresh before they started. A character the console
  cannot hold, such as "✓" or "…", is printed as "?"; the bytes of a file
  name the locale cannot decode are printed as they are. A UTF-8 console and
  a non-raising error mode set with PYTHONIOENCODING are unchanged. This
  fixes console output only: in that locale, writing an HTML report for an
  input file whose name is not ASCII still fails, as in 2.6.1.
• The terminal shows database and input text as it is written. Text in
  square brackets that looked like a style tag was taken as one: a GWAS
  trait or ClinVar condition lost words such as "[bold]", and a closing tag
  that matched nothing, such as "[/x]", ended analyze with a MarkupError
  (exit 1) after the analysis had run. A :name: code became an emoji. The
  same applied to genes, classifications, database versions, a --diff
  baseline's rows, file and path names, sample IDs, the build line read
  from a file, chromosome names and genotypes, in analyze, stats,
  extract, compare, export plink, config and db. A high-value
  no-call line now keeps its reason, such as "[no genotype called]", which
  the terminal used to drop. JSON, HTML and VCF output were not affected.
• The HTML report frames a CADD score as a prediction. A finding's panel
  said "10.4 (top 10% most deleterious)", which next to a ClinVar protective
  finding marked Good read as "bad for you"; it now says "10.4 (top 10%
  predicted deleterious)". The legend adds that CADD predicts how disruptive
  a change is, not whether its effect is good or bad for you: ClinVar and
  the other sources say that, and a protective variant can score high. The
  tiers (10, 20, 30) and the scores are unchanged; the README says the same.
• The console and the HTML report no longer list every call they could not
  check at a high-value site or where ClinVar classifies the variant as
  Pathogenic or Likely pathogenic. 2.6.1 listed them all ahead of the
  findings: 6,625 sites and 14,117 terminal lines on a real 23andMe v5
  file, most of them base calls at 23andMe i probes. They now name a site
  whose rsID is on the high-value list, for any reason, and a call other than
  D/I whose own rsID ClinVar classifies Pathogenic or Likely pathogenic.
  Each other kind of site is one line with its count and why, D/I calls
  at such rsIDs among them. On that v5 file 9 sites are named and the
  terminal report has 553 lines; on a D/I-rich AncestryDNA file, 26 and
  701 lines (2.6.1: 6,609 and 15,228). --show-unchecked (analyze,
  methylation, pharmacogenomics) lists every site on the console as
  before; the HTML report keeps the full list in a collapsed section under
  its banner. The JSON report is unchanged and lists every site
  (ADR-0051).
• The 2.6.1 entry on naming unchecked calls says "the existing counts are
  unchanged". One count did change: the report_warnings entry
  di_calls_not_checked counts the D/I calls the JSON report lists in
  high_value_unchecked (reason di_call), so in 2.6.1 it also counts
  those 2.6.1 began to name, at vendor probes and by a Pathogenic or Likely
  pathogenic row at or near their position (on a real
  23andMe v5 file 376 in 2.6.0, 719 in 2.6.1). The counts that are unchanged
  are input.probe_classes (di_calls_at_plp_rsids among them, 376 on
  that file in both), array_quality, placement and the
  identity_locus_conflict warning (ADR-0051, ADR-0070).