Change or document AR_compound low scoring compound penalty
Opened this issue · 4 comments
The hard coded -6 rank score penalty incurred by AR_comps with only low scoring (arbitrarily hard coded as score 9) variants is not reflected in the rankResults string exported explaining the rank results. It would not hurt to 1) figure out a way to solve the signal-noise of AR_comp without this penalty - it hits very differently in singletons vs trios 2) make it a part of the score model 3) make it appear in the rankResults.
One more case of this https://scout.scilifelab.se/cust002/F0060093/6f7843d6fe355b956031e6aa5664bad7 from @annhamm. In general it is a weakness of the scoring model, and fixing it will allow analysts to "stop" searching at a higher score, giving shorter analysis time. Most false negatives among singletons for score thresholds were found to be caused by this particular flag. A recalibration of trio scores will be needed after changing it, but perhaps not dramatically as AR_comp is a very common occurrence in singletons.
This has now gotten a lot easier in genmod 3.9, thanks to #138. One might say genmod is off the hook even, and move this to a pipeline. Should we say nf-core/raredisease @jemten and @ramprasadn? This of course goes for @fellen31 and @josefinelantz / @mathiasbio as well, but their models are perhaps not so final yet anyway.
Hi @dnil ! Very slow response here, but how difficult would it be to activate this AR_compound function in genmod for cancer? I don't think we're using genmod this way at the moment but only looking at the variants individually. Is this just a manner of turning on some flag for genmod whereupon it would look at variants across genes and increasing the scores of variants if there are other disruptive variants in the same gene?
Thank you for thinking of this - to me, it is slightly off topic here? I answered in one of the original issues instead: Clinical-Genomics/BALSAMIC#596.