DeepWAS: Multivariate genotype-phenotype associations by directly integrating regulatory information using deep...

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Genome-wide association studies (GWAS) identify genetic variants linked to traits but do not directly reveal regulatory mechanisms.
• Functional annotation of genetic variants is typically performed through post hoc analyses.
• Deep learning models can predict variant effects on cell type-specific chromatin features.

Purpose of the Study:

• To introduce DeepWAS, a novel approach integrating variant regulatory effect predictions into a multivariate GWAS framework.
• To directly link single genetic variants to their impact on specific chromatin features within cell types.
• To identify regulatory single nucleotide polymorphisms (dSNPs) associated with multiple sclerosis (MS), major depressive disorder (MDD), and height.

Main Methods:

• DeepWAS integrates deep learning-based predictions of variant regulatory effects into a multivariate GWAS.
• The approach analyzes single variants and their impact on cell type-specific chromatin features.
• Statistical analysis was performed on datasets for MS, MDD, and height, identifying up to 61 dSNPs.

Main Results:

• DeepWAS identified up to 61 regulatory SNPs (dSNPs) associated with MS, MDD, and height.
• These dSNPs were predominantly non-coding and showed nominal significance in classical GWAS.
• DeepWAS demonstrated higher prediction accuracy than classical GWAS for MS-specific dSNPs and showed enrichment in regulatory elements.

Conclusions:

• DeepWAS effectively couples associated variants to their regulatory impact on chromatin features in a cell type-specific manner.
• The method has the potential to generate testable functional hypotheses directly from genotype data.
• DeepWAS offers a powerful tool for advancing the functional interpretation of GWAS results and understanding disease etiology.