clintval / vartovcf / 37780070712

98%

push

github

feat: write FORMAT HICNT for each sample of a tumor-normal pair (#142)

* feat: recover VarDict's hicnt from the SN and ALT depth it prints

VarDict prints SN as hicnt / locnt (hicnt / 0.5 when locnt is 0), a
double formatted to 3 decimals by Java's DecimalFormat, so for a known
total of ALT reads the hicnt that reproduces SN can be found exactly by
bisection. Java's rounding is reproduced on the double's exact binary
value, plus the one case where DecimalFormat rounds the double nearest
0.0005 down. This matches Java 8 on 13.8 million (hicnt, locnt) pairs:
every pair up to 2,000, 2 million random pairs and 7.8 million pairs at
and next to rounding ties; Java 11, 17 and 21 print the same as Java 8
on the full grid.

VarDict's insertion realignment scales the ALT depth, hicnt and locnt
separately, rounding each toward zero, so the ALT depth can be one above
hicnt + locnt; both totals are tried and a count is returned only when
they agree. More than one count fits only with a thousand or more ALT
reads below -q.

* feat: write FORMAT HICNT for each sample of a tumor-normal pair

VarDict counts every ALT read in AD whatever its base quality and only
splits them at -q into hicnt and locnt, which its tumor-normal output
leaves out. Each sample now gets HICNT recovered from its AD[1] and SN:
0 when AD[1] is 0 (VarDict fills a sample without the allele from its
reference reads, SN included), and missing when SN may be the other
sample's, as VarDict copies it to the sample it fills in by subtraction
when it re-calls a long indel on both samples' pooled reads and labels
the call Germline. The tumor-normal header extends the tumor-only HICNT
description with how the count is recovered and when it is missing.

Tumor-only output is unchanged.

* docs: describe the recovered HICNT for tumor-normal input

173 of 175 new or added lines in 3 files covered. (98.86%)

1747 of 1767 relevant lines covered (98.87%)

393360.61 hits per line