Fig. 4

Performance comparison on prioritizing functional variants. Distribution of scores for evaluating five different variant sets, from left to right: variants labeled “benign/likely benign” in ClinVar, variants labeled “pathogenic/likely pathogenic” in ClinVar, randomly selected variants from 1000 Genomes, significant trait-/disease-associated SNPs (cSNPs) from GWAS Catalog and recurrent cancer-related mutations in noncoding regions from COSMIC. a Boxplot for PAFA scores. b Boxplot for Eigen scores. c Boxplot for C scores of CADD v1.3. d Boxplot for Region, TSS, and Unmatched scores of GWAVA. e Boxplot for DANN scores. f Boxplot for coding and noncoding scores of FATHMM-MKL. Area under the curve (AUC) values were calculated for each tool to evaluate their performance on (1) prioritizing pathogenic variants from benign ones, most of which are located in coding regions; (2) prioritizing cSNPs from common variants in 1000 Genomes; and (3) prioritizing noncoding recurrent variants from common variants in 1000 Genomes. As GWAVA and FATHMM-MKL provide more than one score, we display the AUC value with the best performance