CADD

CADD predicts the deleteriousness of human genetic variants by integrating diverse genomic annotations into a unified C score to prioritize functional, deleterious, and pathogenic SNVs, insertions, and deletions.

Key Features:

• Integration of annotations: Combines over 60 genomic features including conservation metrics, functional annotations, pathogenicity signals, regulatory effects, and associations with complex traits into a single score.
• Machine learning classifier: Uses a support vector machine (SVM) trained to distinguish high-frequency human-derived alleles from simulated variants to prioritize functional and deleterious variants.
• Precomputed SNV coverage: Provides precomputed scores for all possible ~8.6 billion human single-nucleotide variants and supports scoring of short insertions and deletions.
• Biological correlation: C scores correlate with allelic diversity, functional annotations, pathogenicity, disease severity, experimentally measured regulatory effects, and GWAS associations.
• Splicing prediction (cadd_phred): The cadd_phred extension incorporates deep neural network (DNN) splicing scores to predict splicing effects beyond canonical donor and acceptor dinucleotides.
• Genome build support: Version 1.4 includes support for the human genome build GRCh38.

Scientific Applications:

• Mendelian disorder variant prioritization: Prioritizes candidate causal variants in severe Mendelian disease analyses.
• GWAS interpretation: Ranks and prioritizes variants underlying genome-wide association study signals using integrated annotations.
• Exome and sequencing analyses: Supports variant effect prediction and prioritization in exome and genome sequencing workflows, including short indels.
• Splicing impact detection: Identifies variants likely to affect splicing, including those outside canonical splice-site dinucleotides.

Methodology:

Integrates >60 genomic annotations, uses an SVM trained on high-frequency human-derived alleles versus simulated variants, provides precomputed scores for ~8.6 billion SNVs and scores for short indels, and incorporates DNN-derived splicing scores in the cadd_phred extension; version 1.4 adds GRCh38 support.