Predrag Radivojac1
1Khoury College of Computer Sciences, Northeastern University, Boston, Massachusetts, USA
predrag [at] northeastern.edu
Abstract
Current ACMG/AMP guidelines for the use of sequence variants for genetic diagnosis and treatment permit the use of in silico predictors as Supporting evidence (PP3 and BP4 criteria). These criteria, however, lack quantitative support and leave clinicians and scientists without standards for applying these criteria, leading to large interpretation variability. To address this challenge, our team built upon previous work and introduced a novel criterion that can be used to calibrate any computational model or any other continuous-scale evidence on any variant type. We used it to estimate score intervals corresponding to the four strengths of evidence for pathogenicity and benignity for fourteen missense variant interpretation tools on a carefully assembled data sets of known pathogenic and benign variants. We found that most tools achieved the Supporting evidence level for both pathogenic and benign classification using newly established data-driven thresholds. Importantly, at appropriate score thresholds, several in silico methods can also provide Moderate and Strong evidence levels for a limited number of variants. Based on these findings, we provided recommendations for quantitative revisions of the PP3 and BP4 criteria within ACMG/AMP guidelines and the future assessment of in silico methods for clinical interpretation.
