Fig. 3: Predictive power of gwRVIS for pathogenic variant classification. | Nature Communications

Fig. 3: Predictive power of gwRVIS for pathogenic variant classification.

From: Prioritizing non-coding regions based on human genomic constraint and sequence context with deep learning

Fig. 3

Mean ROC curves (with fivefold cross-validation) from gwRVIS benchmarking against CADD, phastCons (46-way), phyloP (46 way), and Orion, during ClinVar-pathogenic vs. denovodb-benign variant classification for three noncoding genomic classes: a lincRNAs, b intergenic regions, and c UTRs. The combined performance of gwRVIS with CADD is also shown. ncRVIS is included in the benchmarking of the UTR regions (c), as a robust score specifically designed for the UTR genomic class. One-sided DeLong’s tests have been performed to assess the statistical significance of the differences in predictive power between gwRVIS and the rest of genome-wide scores.

Back to article page