Fig. 3
From: High-throughput deep learning variant effect prediction with Sequence UNET
Generalising Sequence UNET. A Sequence UNET top model ClinVar pathogenicity predictions for the ProteinNet Casp12 test set record TBM#T0865. The wild-type amino acid at each position is outlined. B ROC AUC values comparing VEP performance over the ClinVar test set. Instances tested on the subset of ClinVar with structural data are marked with asterisks (*). C Mean and standard error of Spearman’s rank correlation coefficient between VEP predictions and standardised DMS data [25]. Sequence UNET, ESM-1v, SIFT4G and FoldX predictions were available across all proteins while other tools were only available for human proteins. D ROC AUC values comparing performance of VEPs at classifying known deleterious and neutral S. cerevisiae variants [26]. In B, C and D, the number of variants analysed by each tool is listed on the right