Fig. 4: Comparison of 300K-RNA Top-4* with SpliceAI.
From: SpliceVault predicts the precise nature of variant-associated mis-splicing
a–c, Custom interpretive rules applied to SpliceAI Δ-scores to predict the nature of mis-splicing. Δ-scores below 0.001 were excluded as our applied threshold for no predicted impact on splicing (threshold shown with gray dashed lines). Heights of red lines denote example Δ-scores that predict mis-splicing events according to our rules. a, Single-exon skipping is predicted if both splice sites flanking the exon have donor and acceptor loss Δ-scores above threshold, and double-exon skipping was inferred if the splice site of the upstream or downstream intron also had donor loss or acceptor loss Δ-score above threshold. b, Intron retention was predicted if both splice sites flanking an intron had donor loss and acceptor loss Δ-score above threshold. c, Cryptic activation was predicted by donor gain or acceptor gain Δ-scores above threshold for any unannotated donor or acceptor. d, Example showing SpliceAI predictions of exon skipping and cryptic activation in case number 6. e, Sensitivity and PPV of 300K-RNA and SpliceAI for exon-skipping and cryptic-activation prediction at different thresholds, for the 86/88 variants that can be scored by SpliceAI. Points on the 300K-RNA curve (blue) show metrics when using Top-1*, Top-2*, Top-3*, Top-4*, etc. events as predictions of the nature of mis-splicing. Points on the SpliceAI curve (red) show metrics at Δ-scores that predict the same number of exon skipping and cryptic activation as 300K-RNA Top-1*, Top-2*, Top-3*, Top-4* and so on. f, 300K-RNA and SpliceAI predictions of exon skipping (seen/not seen in RNA studies across 86 cases). g, 300K-RNA and SpliceAI predictions of cryptic splice-site activation (seen/not seen in RNA studies across 86 cases). Dashed lines indicate the threshold of Top-4* and SpliceAI Δ-score ≥ 0.011 identified in e. Black dots, mis-splicing events seen in RNA studies but not meeting the Δ-score threshold of 0.001. An asterisk indicates that filtering to skipping one or two exons and cryptic activation within 600 nt of the annotated splice site. TP, true positives, FN, false negatives, FP, false positives.
