CERENKOV2

CERENKOV2 identifies regulatory single nucleotide polymorphisms (rSNPs) within noncoding genomic regions to prioritize causal variants in GWAS loci.

Key Features:

• rSNP prioritization: Prioritizes regulatory SNPs within noncoding regions highlighted by genome-wide association studies (GWAS).
• Intralocus SNP radius: Defines an intralocus SNP radius as the average data-space distance from a SNP to neighboring SNPs within the same locus.
• Five distance measures: Explores five distinct distance measures to characterize inter-SNP distances and class-based distributional biases.
• Ten distance-based log-likelihood features: Derives ten log-likelihood (likelihood-ratio) features from parametric models of distance distribution differences between rSNPs and control SNPs.
• 248-dimensional feature matrix integration: Integrates the ten distance-based features into an existing 248-dimensional annotation feature matrix.
• Performance metrics: Demonstrates improvements in AUPVR (Area Under the Precision-Recall curve), AUROC (Area Under the Receiver Operating Characteristic curve), and AVGRANK.
• Reference dataset evaluation: Evaluated on the OSU18 reference SNP set comprising 39,083 SNPs.
• Predecessor methods: Builds on CERENKOV, which used 246 annotation features and an xgboost classifier.

Scientific Applications:

• GWAS post-analysis: Prioritizes candidate causal noncoding SNPs within GWAS loci for downstream analysis.
• Regulatory variant discovery: Enhances recognition of regulatory SNPs by incorporating locus-specific geometry in data-space.
• Functional follow-up prioritization: Ranks candidate causal SNPs for targeted functional validation and interpretation of noncoding variant effects.

Methodology:

Defines an intralocus SNP radius as the average data-space distance to neighboring SNPs; computes empirical likelihoods and densities using five distance measures for rSNPs and control SNPs; fits parametric models to derive ten log-likelihood features which are appended to the 248-dimensional feature matrix; evaluates performance on the OSU18 set of 39,083 SNPs; notes that CERENKOV previously used 246 annotation features with an xgboost classifier.