Variant interpretation is central to genomic medicine, but one of its most debated areas is how to use computational evidence. The ACMG/AMP guidelines specify criteria such as PP3 (supporting pathogenicity) and BP4 (supporting benignity), but until recently, applying these criteria consistently across predictive tools has been a challenge. The Calibrated Classification Package in OpenCRAVAT was built to address precisely this need. Computational tools like REVEL, BayesDel, and CADD have long been used to predict the functional impact of genetic variants. However, each tool outputs scores on its own scale, and there has been no universally accepted way to map these scores to ACMG/AMP categories. This lack of calibration has led to inconsistent application of PP3 and BP4 across labs and pipelines. In 2022, the ClinGen Sequence Variant Interpretation (SVI) Working Group published a standardized procedure for calibrating computational predictors to ACMG/AMP evidence strengths (Pejaver et al., AJHG 2022). The Calibrated Classification Package implements this procedure within OpenCRAVAT, providing an open-source, ready-to-use solution for clinical curators and researchers.
What the Package Provides
The package delivers pathogenicity and benignity strength of evidence classifications for seven widely used variant effect predictors. These predictors were rigorously evaluated using ClinVar variants (excluding training variants) to ensure unbiased performance assessment. By embedding calibrated predictors into the OpenCRAVAT ecosystem, this package makes it easier to produce reproducible, evidence-based variant interpretations that are directly linked to community standards. The Calibrated Classification Package moves us closer to harmonizing computational evidence in variant interpretation—helping curators, researchers, and clinicians apply ACMG/AMP criteria with greater precision and reproducibility.
