Supplementary Figure 12: Correlation between classifier performance on human expert-curated ClinVar variants and performance on empirical datasets.
From: Predicting the clinical impact of human mutation with deep neural networks
a, Scatterplot showing the classification accuracy (y axis) on ClinVar variants and on the 10,000 withheld primate variants (x axis) for each of the 20 other classifiers and the PrimateAI network trained with human-only or human + primates data. Shown are the Spearman correlation coefficient rho and associated P value. To limit the evaluation to data that were not used for training the classifiers, we only used ClinVar variants that were added between January 2017 and November 2017, and excluded common human variants from ExAC/gnomAD (>0.1% allele frequency). The ClinVar classification accuracy shown is the average of the true positive and true negative rates, using the threshold where the classifier would predict the same number of pathogenic and benign variants as observed in the ClinVar dataset. b, Scatterplot showing the classification accuracy (y axis) on ClinVar variants and the DDD cases versus controls full dataset (x axis) for each of the 20 other classifiers and the PrimateAI network trained with human-only or human + primates data.
