Enabling genome-wide association testing with multiple diseases and no healthy controls

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• Large-scale whole genome sequencing is often limited to disease cohorts due to cost, creating a lack of healthy controls.
• Existing association studies may yield spurious findings if one disease acts as an unacknowledged control for another.

Purpose of the Study:

• To develop a statistical method for genetic association testing in datasets containing multiple diseases without healthy controls.
• To identify and mitigate confounding associations driven by non-focal diseases within a cohort.

Main Methods:

• Developed genotype-on-phenotype reverse regression (revreg) with a Bayesian spike and slab prior.
• Implemented revreg to perform association testing on datasets with multiple disease phenotypes.
• Designed to flag associations potentially driven by diseases not of primary interest.

Main Results:

• Simulations show revreg achieves 80% power for common variants (1.74 OR) and rare variants (3.73 OR) with low type I error.
• Applied to a whole genome dataset for Chronic Obstructive Pulmonary Disease (COPD), revreg identified six associations likely due to Age-Related Macular Degeneration (AMD).
• In an exome dataset for Asthma, revreg identified and removed genic regions associated with AMD and Rheumatoid Arthritis (RA).

Conclusions:

• Revreg effectively enables genetic association studies in multi-disease cohorts lacking healthy controls.
• The method successfully distinguishes true disease associations from those driven by confounding diseases.
• This approach enhances the accuracy of variant discovery for complex diseases using available large-scale sequencing data.