Calibration and refinement of ACMG/AMP criteria for variant classification with BayesQuantify

Abstract

BackgroundImproving the precision and accuracy of variant classification in clinical genetic testing requires further specification and stratification of the American College of Medical Genetics and Genomics/Association for Molecular Pathology (ACMG/AMP) criteria. While the Clinical Genome Resource (ClinGen) Bayesian framework enables quantitative evidence calibration for selected criteria, standardized tools to optimize evidence thresholds and systematically refine ACMG/AMP criteria remain underdeveloped. MethodsTo address this gap, we developed BayesQuantify, an R package that provides a unified resource for quantifying evidence strength for ACMG/AMP criteria based on the Bayesian framework. BayesQuantify accepts a variant classification file as input and automatically calculates the odds of pathogenicity for each evidence strength, incorporating user-provided prior probabilities of pathogenicity. Through bootstrapping, BayesQuantify generates thresholds by aligning the 95% lower boundary of positive likelihood ratio/local positive likelihood ratio values with the odds of pathogenicity for different evidence levels. Three independent datasets (the ClinVar 2019 dataset, the ClinGen curated dataset, and the PTEN gene dataset) derived from ClinVar, HGMD, and gnomAD were utilized to evaluate the utility of BayesQuantify. ResultsValidation across three independent datasets demonstrates that BayesQuantify delivers objective, consistent refinements for both categorical and continuous ACMG/AMP evidence. Specifically, we replicated the PP3/BP4 thresholds for four computational tools (BayesDel, VEST4, REVEL, and MutPred2) recommended by the ClinGen Sequence Variant Interpretation Working Group using the ClinVar 2019 dataset. Our analysis also indicated that the PM2 criterion should be downgraded from moderate to supporting evidence, aligning with ClinGen recommendations. Importantly, we have established thresholds for supporting, moderate, and strong evidence for in-silico tools using this tool, thereby expanding the application of PP3/BP4 criteria for missense variants in the PTEN gene. ConclusionsBayesQuantify is an accessible and user-friendly resource that enhances the rigor and reproducibility of ACMG/AMP criteria application. By facilitating evidence-based stratification and threshold optimization, the tool strengthens variant classification workflows, offering immediate value to clinical genetic testing laboratories and research communities. The package is freely available at https://github.com/liusihan/BayesQuantify.