Replace ClinVar whitelist with optional ClinVar-based scoring

[julesjacobsen]

The ClinVar whitelist option (default = true) will set all P/LP 2*+ variants found in a VCF file to have a score of 1.0.

The issue with this is that, while it works well for pathogenic variants, it is useless for benign variants.

Another issue is that there is no longer a distinction between P and LP. The ACMG guidelines state that there should be a >=0.99 for P and 0.90-0.98 probability of pathogenicity for these categories and a similar range for B/LB.

Converting these probabilities to variant scores which override the default implementation will allow us to keep the boost to the known pathogenic variants and also de-prioritise known benign variants, some of which would otherwise score rather highly.

The implementation issue here is should this be an optional setting which can be set to false for use in internal benchmarks or would this be a case of YAGNI? Also, how would this interact with the whitelist implementation? Would this need to be updated so that the variants on the list should be given a P/LP/LB/B label?

[damiansm]

Sounds like a good idea. I guess we could produce a new whitelist file along these lines for all 2* variants chr pos ref alt clinvar_category override_variant_score var1 P 1 var2 LP 0.95 var3 VUS - var 4 LB 0.05 var5 B 0 So for P, LP, LB and B override the usual variant score with these values based on clinvar evidence. For VUS just rely on the Exomiser variant score. Need to think if we want to whitelist LB and B variants that would otherwise be filtered out - I think no, otherwise we will be swamped by loads of common variants.

…

On Wed, Sep 6, 2023 at 8:45 PM Jules Jacobsen ***@***.***> wrote: The ClinVar whitelist option (default = true) will set all P/LP 2*+ variants found in a VCF file to have a score of 1.0. The issue with this is that, while it works well for pathogenic variants, it is useless for benign variants. Another issue is that there is no longer a distinction between P and LP. The ACMG guidelines state that there should be a >=0.99 for P and 0.90-0.98 probability of pathogenicity for these categories and a similar range for B/LB. Converting these probabilities to variant scores which override the default implementation will allow us to keep the boost to the known pathogenic variants and also de-prioritise known benign variants, some of which would otherwise score rather highly. The implementation issue here is should this be an optional setting which can be set to false for use in internal benchmarks or would this be a case of YAGNI. Also, how would this interact with the whitelist implementation? Would this need to be updated so that the variants on the list should be given a P/LP/LB/B label? — Reply to this email directly, view it on GitHub <#515>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABHO4PA2JNGNBXQYWETPMYTXZDHERANCNFSM6AAAAAA4N2UYNU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[julesjacobsen]

Yes, that's the idea. There are these issues to address:

1. Do we need to define the override_variant_score column? This should be a function of the clinvar_category as defined by the ClinGen SVI group, otherwise users will need to add this themselves and add potentially inconsistent scores.
2. The current ClinVar whitelist will be replaced with an implementation which automatically filters the ClinVar variant store (see Enable more frequent ClinVar data updates in Exomiser database #501) and populates the whitelist with 2*+ P/LP variants. This whitelist implementation will need to be altered slightly to return an AcmgCategory (P/LP/LB/B) instead of true/false and this added to the VariantEvaluation.

The (now inaccurately named) whitelist can still be toggled on and off using the exomiser.${assembly}.use-clinvar-whitelist option.