Bayesian Probability-Based Classification for CYP2D6 Copy Number Determination by Digital PCR

Abstract

Background: Accurate determination of CYP2D6 copy number variation (CNV) is essential for reliable phenotype assignment in pharmacogenetics. Although digital PCR (dPCR) is widely used for CNV analysis, standardized approaches for selecting and probabilistically evaluating decision intervals for converting quantitative measurements into categorical copy number assignments remain limited. Methods: The Absolute Q dPCR platform was evaluated using a Bayesian classification model to assess an empirically selected classification interval for CYP2D6 copy number assignment. Expected distributions for each copy number category were derived from characterized reference materials. Based on the observed distributions and the manufacturer-recommended acceptance interval, an interval was empirically selected and subsequently evaluated using Bayesian posterior probabilities. Analytical accuracy was evaluated by comparing dPCR-derived copy number assignments with reference diplotypes using two multiplex dPCR assays targeting distinct CYP2D6 regions. Results: Evaluation of the empirically selected ±0.35-copy interval yielded posterior probabilities of >99.9% for measurements within the corresponding copy number intervals. Measurements outside these intervals were classified as nonconclusive. The multiplex assay targeting CYP2D6 intron 2, exon 9, and the 5′UTR showed perfect concordance with reference diplotypes, whereas the assay targeting intron 2, exon 9, and intron 6 showed lower concordance because it failed to detect *68 hybrid alleles. Conclusions: This Bayesian probability-based framework provides a probabilistic evaluation of an empirically selected classification interval for converting quantitative dPCR measurements into categorical CYP2D6 copy number assignments. The empirically selected ±0.35-copy interval yielded high posterior probabilities while providing an explicit operational rule for classifying measurements outside the interval as nonconclusive, supporting analytical validation and routine clinical implementation.

The authors' abstract, as published at the source. BioMedInformatics, 2026 · DOI ↗