Use of Clinical Evidence in VarSome’s ACMG implementation

Rules PP5 and BP6 are still used in Saphetor’s ACMG classifier, although ClinGen has recommended that they should be phased out in order to avoid double-counting the same evidence or to provide stricter evidence criteria for research.

Saphetor’s ACMG classifier implements as many of the ACMG guidelines as is technically possible, however the guidelines were written for humans with a lot of background knowledge, domain experience, and a familiarity with the relevant scientific literature. We calibrate our system with evidence from external databases like ClinVar and UniProt (we refer to these as "clinical evidence") disabled, in order to ensure that the classifications match those from clinical databases as accurately as possible, thus proving that the system will accurately classify novel variants.

However, there are many variants that cannot be automatically classified: the evidence falls short, or the in-silico predictors may not be available or correct. We have therefore chosen to continue applying rules PP5 and BP6, and assign various evidence strengths based on the credibility of the source. ClinVar stars are very helpful in this respect; we assign strength Very Strong to 3-star clinical evidence reviewed by an expert panel, for example.

Overall, leveraging clinical evidence in this way aims to ensure that any reported pathogenic evidence can be easily identified, even if the automated classification is VUS or Benign. In VarSome Clinical it is simple to filter on rule PP5.

Lastly, the ACMG Classifier in the VarSome front-end allows users to disable “clinical evidence” and re-classify the variant without it. In many instances, the additional evidence provided by databases such as ClinVar is not necessary, though it provides additional validation.