Fig. 3
From: Benchmarking splice variant prediction algorithms using massively parallel splicing assays
Splice effect predictors’ classification performance on benchmark variants. A Precision-recall curves showing algorithms’ performance distinguishing SDVs and splicing-neutral variants in each dataset. B Precision-recall curves of tools’ performance differentiating SDVs and splice neutral variants in exons (left) and introns (right). C Top panel: tally, for each algorithm, of the number of individual datasets and variant classes (defined as in Fig. 1B) for which that algorithm had the highest prAUC or was within the 95% confidence interval of the best performing tool. Bottom panel: signed difference between the best performing tool’s prAUC and a given tool’s prAUC; each dot corresponds to an individual dataset or variant class