clintval / chaff / 37936089633

97%

push

github

feat: start end repair fill-in on a duplex consensus from a 5 bp scale (#7)

* feat: start end repair fill-in on a duplex consensus from a 3 bp scale

A duplex consensus removes fill-in errors, which sit on one strand, so on it end repair fill-in already measures each template from the nearer end and scores the changes both strands agree on near the fragment ends. Those sit within a few bases of the end, but the decay started from, and was shrunk toward, the 15 bp fill-in default, and a library holds too few such calls to move it, so a real call with molecules 10 bases in scored almost like one at the end.

The decay now starts from 3 bp when most of the filter's calls are a duplex consensus (aD and bD both above zero); reads and simplex consensus keep 15 bp, and a fixed --end-repair-fill-in-distance still wins.

* docs(readme): show end repair fill-in as the duplex fragment-end filter

On a duplex consensus the filter measures each template from the nearer end and scores changes of any class that crowd the fragment ends, which a caller's read-position filter cannot tell from the clip points that overlap clipping leaves mid-fragment. The README now says so where it sent Duplex Sequencing users to copied damage alone, and runs the examples' reads again as a duplex consensus: tests/data/duplex.bam holds the same records as tumor.bam, each tagged aD:i:1 and bD:i:1.

* feat: start end repair fill-in on a duplex consensus from 5 bp

At 5 bp and a threshold of 0.5 the filter catches nearly every fragment-end call a read-position filter does, at under half its cost to real variants, where 3 bp caught about three quarters. The README example's posteriors and learned scale follow, and the duplex test's call at 100 now scores 0.0012, just above that test's threshold of 0.001.

14 of 14 new or added lines in 2 files covered. (100.0%)

2220 of 2293 relevant lines covered (96.82%)

395826.06 hits per line