clintval / chaff / 38336991894

97%

push

github

fix!: fit the copied-damage dispersion by maximum likelihood, shared across strata (#16)

* fix!: fit the copied-damage dispersion by maximum likelihood

The single-strand dispersion matched the positions expected with no change to those observed, which a Poisson and a negative binomial separate only at second order at these rates, so the shape swung with noise and was finite in about half of Poisson libraries.

Fit the shape's inverse by maximum likelihood over every position's count, the rate profiled at each shape, and keep a Poisson unless a likelihood ratio test rejects it at the 1% level.

BREAKING CHANGE: CDAP can move on a duplex BAM with single-strand consensus.

* fix!: share the copied-damage dispersion across a library's strata

A stratum holds too few single-strand changes for the likelihood ratio test to find a gamma shape of 2 to 10, so the fit fell back to a Poisson and the chance prior ran low. A stratum whose own changes cannot tell now takes the shape its library's strata share, each at its own rate, fitted over the strata whose own shapes do not reject it at the 1% level.

Every shape must also beat a Poisson without the one position that favours it most, so a single clustered position no longer widens a stratum, nor through the shared fit the others.

BREAKING CHANGE: CDAP, chance_fraction and chance_excess can move on a duplex BAM with single-strand consensus wherever a stratum's fitted shape changes; no option or column changes.

158 of 158 new or added lines in 2 files covered. (100.0%)

2452 of 2523 relevant lines covered (97.19%)

6688504.28 hits per line