A machine-learning framework to characterize functional disease architectures and prioritize disease variants

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) are crucial for understanding disease functional architecture and prioritizing causal variants.
• Accurate modeling of disease effect sizes is essential for advancing genetic research.

Purpose of the Study:

• Introduce the variant-to-disease (V2D) framework, a machine learning approach for modeling disease effect sizes.
• Leverage posterior estimates from genome-wide fine-mapping and functional annotations.
• Enhance the prioritization of potentially causal variants.

Main Methods:

• Developed the variant-to-disease (V2D) framework using machine learning algorithms.
• Applied linear trees to model heritability and functional enrichment across 15 UK Biobank traits.
• Utilized neural networks to generate GWAS prioritization scores.

Main Results:

• V2D framework provides reliable estimates of heritability (h²).
• Identified non-linear relationships between constraint and regulatory annotations, highlighting constrained regulatory variants.
• Developed GWAS prioritization scores enriched in common variant h² (20.6 ± 0.7x for top 1%), outperforming existing methods.

Conclusions:

• The V2D framework offers a robust method for modeling disease effect sizes and functional architecture.
• Constrained regulatory variants are a major component of disease functional architecture.
• V2D-derived GWAS prioritization scores improve variant discovery and are transferable across datasets.