Fig. 2
From: Benchmarking splice variant prediction algorithms using massively parallel splicing assays
Agreement between predictors and experiments varies by gene region. A Splicing effect tracks at alternate isoforms beta and alpha from a published MPSA [57] of POU1F1 exon 2. Upper panel tracks (gray background) show MPSA-measured percent of exon skipping, beta exon inclusion, and other (non-alpha/beta/skip) isoform usage; lower tracks show scores from bioinformatic predictors by position. Each lollipop denotes one variant, shaded by effect in MPSA (gray: neutral, colors: SDVs, shaded by mutant base). The exon and flanking introns are split by region. B Heatmap showing concordance between each algorithm’s binary classification of variants as SDV/neutral versus those of the MPSAs. For each algorithm, the score threshold that maximizes Youden’s J across the full POU1F1 dataset is used. Concordance is shown per algorithm (row) vs each region (column) at that score threshold. Regions with <10 scored variants are omitted (“X” symbols)