Fig. 3: Classification performance for detecting sQTLs in the Icelandic RNA-Seq using Transformer-45k and pre-trained SpliceAI-10k. | Communications Biology

Fig. 3: Classification performance for detecting sQTLs in the Icelandic RNA-Seq using Transformer-45k and pre-trained SpliceAI-10k.

From: Transformers significantly improve splice site prediction

Fig. 3

a PR-AUC plotted against maximum distance from an sQTL to the closest splice site annotation. b Precision-recall curve for sQTLs determined to be splice-disrupting or splice-creating. c Precision-recall curve for 35,464 pathogenic splice variants in ClinVar. d A scatter plot showing the distribution of delta scores for non-splicing variants (n = 40,528), benign splice variants (n = 1001), and pathogenic splice variants (n = 35,464).

Back to article page