Fig. 4

Sensitivity and specificity trade-off as cutoff and threshold vary exon-level classification performance of geneCNV model on nine Coriell samples, after hyperparameter training on 38 volunteer samples. Among the nine Coriell test samples here, there were a total of 77 affected exons and 634 unaffected exons, used in calculating sensitivity and specificity respectively. a shows the effects of varying the credible interval cutoff on the proportion of callable targets, true positives (sensitivity), and true negatives (specificity) for this test set. Exons where the highest-density interval of the chosen cutoff size spans two copy number states are given an “uncertain” call and not included in subsequent sensitivity and specificity analysis. b shows the effects of varying the threshold for abnormal copy number state probability (as defined in Methods) on sensitivity and specificity. Note that every exon is given a copy number call using this schema