Statistic TumorSampleAltAlleleFraction is truncated

[tdelhomme]

Dear all,

This is not really an issue about the Strelka2 variant caller, but more a question about a particular statistic used in the algorithm:

I simply run Strelka2 (somatic mode) on a TCGA WES bam file (downsampled for an external purpose), and ask to output the variant statistics used in the machine learning algorithm (--reportEVSFeatures).

The thing I don't understand is: why the variant allelic fraction of somatic calls (PASS and not PASS) are truncated? Maximum is 50%, see png file attached.

Does the algorithm re-scale this value?

Thanks in advance,

Tiffany

[tdelhomme]

Note: I think it is not "truncated" but more re-scaled, i.e.

if TumorSampleAltAlleleFraction > 0.5 then TumorSampleAltAlleleFraction = 0.5

[tdelhomme]

I found this piece of code which re-scales the AF to prevent for LOH regions. Does this have an influence on the interpretation of the final EVS score? See this post