Inner cutoffs under calling methods

[anasKh79]

The inner cutoffs mentioned in the paragraph seems to be inconsistent with the values in the table in the documentation.

[etal]

It's different math. The table shows rounding to the nearest integer copy number assuming 100% purity, e.g. a region with a single-copy gain in an otherwise diploid genome will have a log2 ratio of about log(3/2) = 0.58. Therefore, what does an observed log2 ratio of 0.3 imply? It's closer to 0.58 than to 0.0, so CNVkit rounds it to an integer copy number of 3 in a diploid genome (3/2). Put differently, the cutoff for rounding up or down is log(2.5/2) = 0.322 and at that point the signal is rough enough that we only trust it to 1 decimal place. That's the logic you see in the table. The call command allows you to input the estimated purity and adjust before performing this math, so the cutoffs can assume 100% purity.

The --threshold default is more pragmatic. Solid tumor samples are never 100% clonal, they're more often in the range of 30-50% purity, and purity estimates are noisy, too. False negatives are worse than false positives at this point in the overall flow (FP can be discarded by clinical geneticist reviewers downstream, but FN can't be recovered), so we default to an uninformed guess of 30% purity to ensure the pipeline does the right thing when purity information is not available. Obviously, having a better purity estimate will help reduce FP or FN rates.