DeepWAS

DeepWAS integrates deep learning–predicted regulatory effects of single nucleotide polymorphisms into genome-wide association studies (GWAS) to prioritize regulatory variants and link them to cell type–specific chromatin features.

Key Features:

• Deep learning predictions: Uses deep learning models to predict regulatory effects of individual variants across various cell types and chromatin features.
• Variant-level chromatin impact: Couples single-variant predictions to their predicted impact on chromatin features to connect SNPs with regulatory mechanisms.
• dSNP identification: Identifies regulatory SNPs (dSNPs), reporting up to 61 dSNPs in applications including multiple sclerosis (MS), major depressive disorder (MDD), and height.
• Non-coding variant prioritization: Prioritizes variants predominantly residing in non-coding regions that show at least nominal significance in classical GWAS settings.
• Comparative predictive performance: Reports improved predictive performance versus conventional methods, e.g., 91% of genome-wide significant MS-specific dSNPs show higher performance with DeepWAS.
• Regulatory QTL enrichment: Shows enrichment of identified dSNPs in expression and methylation quantitative trait loci (eQTLs and mQTLs).
• Hypothesis generation: Enables generation of testable functional hypotheses from genotype data by linking variants to predicted regulatory effects.

Scientific Applications:

• Neuropsychiatric and autoimmune genetics: Identification and prioritization of regulatory variants associated with multiple sclerosis (MS) and major depressive disorder (MDD).
• Human trait genetics: Prioritization of regulatory variants associated with complex traits such as height.
• Functional follow-up candidate selection: Prioritizing non-coding SNPs for experimental validation based on predicted regulatory impact and GWAS association.
• Integrative regulatory analysis: Linking GWAS signals to molecular regulatory mechanisms through eQTL and mQTL enrichment analyses.

Methodology:

DeepWAS applies deep learning models to predict per-variant regulatory effects across cell types, couples these predictions with GWAS associations to identify regulatory SNPs (dSNPs), and assesses enrichment in eQTLs and mQTLs and comparative predictive performance versus conventional GWAS methods.