Fixed
- A two-ALT site written over two or more VCF records no longer hides the
ClinVar and GWAS Catalog findings 2.6.3 reported. 2.7.0 read such
records as one call only when they were the exact complementary pair
with the same ID (1/0+0/1, or1|0+0|1in one phase set), and
gave every other shape no call at all. Now:- The pair is also
1/.+./1(and1|.+.|1in one phase set),
asvt decomposeandbcftools --multi-overlaps .write a1/2
call, and a missing, positional or other ID never keeps two records
from pairing: only two different rsIDs do. The 1000 Genomes files,
which write one positional ID per allele, now pair: HG00096's eight
split sites on chromosome 22 give their eight calls. The pair's rsID
is decided per ALT: each record keeps the rsID it names, and a record
naming none has its ALT looked up alone by position. Two different
rsIDs (both found by position, or one written and one found) make the
records no pair, as if they named those rsIDs: each ALT is a half call
under its own rsID. Exactly one rsID names the full call, even where
the other ALT belongs to no known rsID. Either way the GWAS Catalog
finding 2.6.3 gave is back, andextractfinds the rsID. ClinVar
then shows two half calls (A/.,C/.) at a two-rsID site instead
of the person'sAC. - Records that agree but do not make one genotype (
0/1+0/1
unphased, different phase sets, a half call beside a heterozygote, two
different rsIDs, a passing record beside a filtered, low-quality or
uncalled one) give each passing record's allele as a half call
(A/.): it goes only to the sources that read one allele (ClinVar and
the GWAS Catalog, where they run), not to ClinPGx or SNPedia. A
rejected sibling does not cancel a passing record's ALT; the site still
follows the half-call rule, so the passing record's full genotype is
not kept. - Records that contradict each other (more than two ALT copies, a
haploid beside another call, or one haplotype carrying two ALTs) still
give no call, and the site is now named inhigh_value_unchecked
(split_site): under each high-value rsID it stands for, written by
its records or placed at its position with its REF (whatever IDs or
ALTs its records carry, as for a single record), and where ClinVar
holds a Pathogenic or Likely pathogenic row at its position for an ALT
the records carry, whatever rsIDs the records name. A row without an
rsID is given as evidence by its locus and gene. split_site_callsstill counts every record of a site that is not
the pair; its warning now says that where such records agree, their
alleles go only to the sources that read one allele. A pair split only
because its ALTs belong to two different rsIDs has its own sentence:
the genotype there is known, and the alleles are assessed apart. The
half-call note counts standalone half calls left after the split-site
decision; the split-site alleles have their own count and line. Both
are given only to the sources that read one allele, where those
sources run.
docs/formats.md describes the rule; ADR-0062 is amended.
- The pair is also
extractshows the genotypeanalyzereports at a site written over
several VCF records. 2.7.0'sextractread each record on its own, so
forT>A 1/0+T>C 0/1it showed the first record'sA/Twhere
analyzereportsAC. It now reads the exact pair asanalyzedoes,
on the sequential and the indexed (tabix) path: one full call. Records
that contradict each other give no genotype; at a site whose records
only agree,extractshows one of their half calls, whileanalyze
reports each.compare,statsandexport plinkstill read every
record on its own.extractno longer shows-/-whereanalyzereports a split site's
half call. A rejected record (FILTER or the sample's FT) written before
the passing one gave a no-call that won on file order, soT>A 1/0
rejected +T>C 0/1passing showed-/-instead ofC/-. At a
placement, a call with a known allele, full or half, now beats a full
no-call; placement still comes first, and file order breaks other ties.- Allelix reads its own annotated VCF again.
--vcf-outwrote the rsID
it recovered for an unnamed split pair on the first record only, so a
second run saw two IDs and gave no findings. Both records of a merged
pair now carry its rsID and their ownALLELIX_ORIGINAL_ID. --diffagainst a baseline from the other genome build no longer
reports every finding as removed and new. 2.7.0 paired findings by
position as well as variant, and positions differ between GRCh37 and
GRCh38, so a chip baseline against a whole-genome run of the same person
looked like a complete change. When the builds differ, findings named by
an rsID are now paired by variant and allele, real additions, removals
and reclassifications are still reported, and the run prints one note
naming both builds. A finding known only by its position (a ClinVar
variant without an rsID) can still show as removed and new across
builds. A baseline whose build is the same, missing or unknown is paired
by position as before.- The HTML report's ClinVar review stars follow ClinVar's rubric. They
were guessed from the wording of the review status, so a submission
reviewed by an expert panel showed four stars instead of three, and
other statuses came out one to four stars off. Each status ClinVar
defines now has its own star count: four for a practice guideline,
three for an expert panel, two for multiple submitters with no
conflicts (or, for a somatic clinical impact, multiple submitters), one
for a single submitter or conflicting classifications, and none for the
rest. A status ClinVar does not define is shown as its
text, with no stars.
The JSON report, the console, scores and the zero-star rule (ADR-0052)
were never affected. - An HTML report with no findings has a working dark-mode button, and no
longer shows a search box and filter buttons that did nothing. The page
script stopped before wiring any control when there was no findings
table. - The closed detail panel of the HTML report no longer leaves a faint
shadow strip down the window's right edge; the panel casts its shadow
only while it is open. - A long cache path in the HTML report's "Sources not used" list wraps
instead of making the page scroll sideways on a phone. - A position above 2^63 - 1, a VCF POS below 1 or a negative array
position is a malformed row. A VCF record or an array row at a position
of 2^63 or more endedanalyze,pharmacogenomics,methylation,
extract,statsandexport plinkwith a Python traceback
(OverflowError), and a VCF POS of 0 or below was read as a variant.
Every positioned reader now skips such a row with a warning, and the VCF
reader counts it inmalformed_rows, as it does a POS that is not a
number. Positions 1 to 2^63 - 1 are read as before, and an rsID list
without positions is unchanged. An array row at position 0 is still read
as before, as an rsID call the vendor did not place, with no warning: the
sources that match by rsID (ClinPGx, the GWAS Catalog, SNPedia) still
report it. --vcf-outon a file system without hard links (FAT, exFAT, some
network shares) is refused at start-up. 2.7.0 refused it only after the
whole analysis, with the JSON or HTML report already written, so a
whole-genome run on an external drive could end this way after an hour.
The run now makes a test link in the destination's folder first and
stops there, with the same message and nothing left in the folder.
Changed
- Reading a VCF for
analyze,methylationandpharmacogenomicsis
faster again. 2.7.0's split-site reading checked every single-base
record for a second record at its position, adding about 40% to the
parse. It now checks only when the next record has the same chromosome
and position, with the same results: on a 600,000-record VCF the
parse's extra cost fell from 43-46% to 0-17% (two rounds each). - JSON reports use schema 15 (2.7.0 wrote 14), and
--diffreads
schemas 1 to 15. Schema 15 addsinput.vcf_quality.split_site_half_calls
and thehigh_value_uncheckedreasonsplit_site.half_callsnow
counts standalone half calls left after the split-site decision; unlike
schema 14, it excludes half calls belonging to a split site. A merged
exact pair contributes to neither count. Split alleles eligible for
half-call assessment have the separatesplit_site_half_callscount.