Variant effect predictors are shaping the ClinVar classifications used to train and evaluate them

Rare missense variants from a major ClinVar submitter show likely benign classifications shifting markedly toward more benign BayesDel_noAF scores, with a sharp step at the calibrated moderate-evidence threshold.

• Why it matters: This matters because VEP predictions can influence the ClinVar classifications later used to train, benchmark, and calibrate VEPs, creating classification-level circularity that undermines independence.
• What they did: The study analyzed rare missense variants from a major submitter and found a step change at the calibrated moderate-evidence threshold of BayesDel_noAF, extending to nearly all 52 predictors examined.
• The result: Because computationally informed classifications cannot be treated as independent ground truth and temporal holdouts cannot remove the circularity, the authors recommend standardized recording of evidence criteria and increased use of classification-independent benchmarks.