clintval / vartovcf / 39901213645

98%

push

github

feat: decide a leaky MNV's partial rows by their reads with --reads (#160)

* build: add streampile, noodles and bstr to read BAMs

streampile piles reads up at positions and calls a template's mates into
one base, noodles opens an indexed BAM and queries a region, and bstr
holds the sort order the pileup builder reads from the header.

* feat: decide a leaky MNV's partial rows by their reads with --reads

With --reads naming the indexed BAM VarDict read for each sample, a
partial row is decided by its reads instead of its count: the reads over
the MNV's span are piled up, each read's walk replays VarDict's MNV
growth so it feeds the row VarDict counted it toward, and each template
feeding the partial is read at the MNV's other changed positions. A
template with the MNV's ALT there at -q or more is the MNV's molecule
and its reads move; one with another base at -q or more is another
allele's and its reads stay; one that proves nothing follows the row's
other reads, moving only where the MNV has reads in the sample and no
template of the row there is another allele's.

A partial whose every read moves in every sample is folded whole as
before; one with reads left is written with only them, recomputed from
those reads, with FORMAT MNV_MOVED_READS. Five options mirror the
VarDict run: --reads-base-quality (-q), --reads-min-mapq (-Q),
--reads-exclude-flags (-F), --reads-max-mismatches (-m) and
--reads-mnv-gap (-X). The MNV_PARTIALS header line now says how a
partial is recognized by counts or by reads, which changes three
fixtures' header line.

* test: cover the read-decided fold end to end with built BAMs

The BAMs are built in the tests with streampile's SamBuilder and
indexed with noodles, so no binary fixture is committed: a mate that
proves the MNV molecule, a no-call at the other changed base, the
reference there, a no-call at the first base, an SNV with another ALT,
and a tumor-normal pair where one sample moves its reads and the other
keeps them;... (continued)

1560 of 1583 new or added lines in 5 files covered. (98.55%)

5780 of 5884 relevant lines covered (98.23%)

122598.52 hits per line