Unified meta regression models for rare variant association studies

Area of Science:

• Genetics and Genomics
• Computational Biology
• Drug Discovery

Background:

• Rare variant association studies (RVAS) are crucial for understanding complex traits, aiding drug discovery and diagnostics.
• Predictive models like AlphaMissense and constraint metrics help identify deleterious and functionally important genetic variants.
• Loss-of-function (LoF) variants offer clear insights into downstream functional consequences.

Purpose of the Study:

• To develop a unified meta-regression model integrating variant pathogenicity, constraint, and type (LoF/missense) for association analysis.
• To model observed effect sizes and uncertainties from single-variant genetic analyses.
• To characterize gene discoveries by the contribution of constrained sites, predicted pathogenic sites, and variant types.

Main Methods:

• Developed a unified meta-regression model incorporating AlphaMissense pathogenicity, constraint probabilities, and LoF/missense indicators.
• Applied the model to 1,144 UK Biobank continuous phenotypes using Genebass single-variant summary statistics.
• Validated findings using the AllofUS cohort.

Main Results:

• The unified model successfully integrated diverse variant features to analyze genetic associations across numerous phenotypes.
• Characterizations of gene discoveries regarding constrained sites, predicted pathogenic sites, and variant types were generated.
• Results are publicly accessible via the Global Biobank Engine.

Conclusions:

• The unified meta-regression approach provides a robust framework for interpreting rare variant associations in complex traits.
• Integrating multiple variant-level features improves the power to detect and characterize genotype-phenotype relationships.
• This work facilitates enhanced drug discovery and diagnostic applications by providing comprehensive variant annotations.