The PS4-likelihood ratio calculator: flexible allocation of evidence weighting for case-control data in variant classification

Abstract

Background: The 2015 American College of Medical Genetics/Association of Molecular Pathology (ACMG/AMP) variant classification framework specifies that case-control observations can be scored as 'strong' evidence (PS4) towards pathogenicity.

Methods: We developed the PS4-likelihood ratio calculator (PS4-LRCalc) for quantitative evidence assignment based on the observed variant frequencies in cases and controls. Binomial likelihoods are computed for two models, each defined by prespecified OR thresholds. Model 1 represents the hypothesis of association between variant and phenotype (eg, OR≥5) and model 2 represents the hypothesis of non-association (eg, OR≤1).

Results: PS4-LRCalc enables continuous quantitation of evidence for variant classification expressed as a likelihood ratio (LR), which can be log-converted into log LR (evidence points). Using PS4-LRCalc, observed data can be used to quantify evidence towards either pathogenicity or benignity. Variants can also be evaluated against models of different penetrance. The approach is applicable to balanced data sets generated for more common phenotypes and smaller data sets more typical in very rare disease variant evaluation.

Conclusion: PS4-LRCalc enables flexible evidence quantitation on a continuous scale for observed case-control data. The converted LR is amenable to incorporation into the now widely used 2018 updated Bayesian ACMG/AMP framework.

Keywords: Genetic Testing; Genetic Variation; Genetics; Genetics, Population.

Figure 1. Overview of PS4-LRCalc framework for flexible PS4 application. (1) For a given set of case-control variant observations, the expected background odds of selecting a case among variant carriers are calculated using the equivalent observed odds among non-carriers. (2) The background odds are then scaled according to the ORs of association and non-association and converted to an expected probability of selecting a case among variant carriers under each hypothesis. (3) Variant observations in cases and controls are then modelled using a binomial likelihood function, which evaluates the likelihood that a given probability p of case selection would generate the observed data (k variant observations in cases across n total observations); note that these probabilities directly convert to odds values, which in turn generate a continuum of ORs across all possible values of p. (4) The likelihood ratio (LR) towards pathogenicity is determined by quantifying the total likelihood of association (L(Assoc); red area under curve) and dividing it by the total likelihood of non-association (L(NonAssoc); blue area under curve). The LR can then be converted to Tavtigian exponent points (EPs) by taking its log (to base 2.08). PS4-LRCalc, PS4-likelihood ratio calculator.

Figure 2. Comparison of applicable strength for the PS4 criterion between the 2015 ACMG/AMP guidelines and PS4-LRCalc for selected case-control scenarios. Counts of carriers (V⁺) and non-carriers (V⁻) of a variant between cases and controls are illustrated for exemplar scenarios indicative of (A–C) pathogenicity and (D–F) benignity shown in tables2 3. In the PS4-LRCalc approach, the distribution of variant carrier observations between cases and controls is modelled using a binomial likelihood curve, in which the likelihood of association (OR≥5, red) is divided by the likelihood of non-association (OR≤1, blue) to generate a likelihood ratio (LR) towards pathogenicity, that can then be converted to Tavtigian exponent points (EPs) towards pathogenicity or benignity and a corresponding applicable evidence strength under the 2015 ACMG/AMP framework. Notably, our approach allows assignation of EPs for scenarios that may not fulfil the existing ACMG/AMP PS4 guidance, including in support of benignity. ACMG/AMP, American College of Medical Genetics/Association of Molecular Pathology; PS4-LRCalc, PS4-likelihood ratio calculator.