Binary classification metrics

Prediction tools may classify variants as benign or pathogenic, but may also fail to reach a classification or classify a variant as VUS. Because of these three outcome states, binary classification metrics must be used with caution. We define sensitivity as the number of detected pathogenic variants (true positives) over the total number of pathogenic variants, which includes true positives, false negatives (pathogenic variants misclassified as benign), and pathogenic variants that were otherwise “missed,” i.e. classified as VUS or not classified at all. Therefore, Sensitivity = TruePositive/(TruePositive + FalseNegative + MissedPositive). We applied the same definition for specificity and define it as: Specificity = TrueNegative/(TrueNegative + FalsePositive + MissedNegative). Following this line, accuracy is then defined as (TruePositive + TrueNegative)/(TruePositive + TrueNegative + FalsePositive + FalseNegative + MissedPositive + MissedNegative).