Closed dnil closed 8 years ago
We have reasoned that the inheritance patterns should be used as just another way of filtering the variants. If they are already annotated with deleteriousness and/or frequency it should be rather easy to filter on these variables at once? I agree that the AR_comp model is weak when there is only one individual but there is no obvious way of handling this.
There is a --strict
flag that will only annotate models if they have "proof", this will reduce the number significantly in for example a trio.
But I will be happy to discuss this in more detail, pleas elaborate on how your idea should be done.
Closing this for now
Ah, sorry, left that one hanging.
Not urgent in any way, but I still feel one could go further and make the tool even better.
The key thing is “filter … at once”. As is, the inheritance model and other filter criteria is evaluated separately. That is fine for independent events/variants. But compound-ness is a cross-term between variants.
What I would like to be able to do is to see only AR_comp for those variants that have a comp partner left after other filtering; e.g. rarity and consequence types in a certain severity range.
The AR_comp model is very inclusive, and so renders much of filtering on this model difficult. It would be useful to be able to tune it to a more stringent approach, akin to what the combined rank score models tries to achieve - or a manual assessment gives. Several solutions would be possible.
Perhaps introducing a different model AR_comp_strict that would be valid only for rare- or high ranking deleterious variants? Neither a manual operator, nor eg ASMG guidelines would consider a common variant valid as disease causing under an AR compound model except when experimental evidence warrants it (e.g. as in Bardet-Biedel or in a select few other disorders).