Second-order group knockoffs with applications to genome-wide association studies.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Conditional testing with knockoff framework identifies unique explanatory variables and controls false discovery rate.
• Genome-wide association studies (GWAS) aim to find genetic variants influencing medical traits.

Purpose of the Study:

• To expand algorithms and software for group knockoffs, addressing challenges in distinguishing correlated regressors in GWAS.
• To develop and implement efficient second-order knockoff methods suitable for GWAS data.

Main Methods:

• Developed novel group knockoff algorithms, focusing on second-order knockoffs.
• Introduced correlation matrix approximations for computational savings in GWAS.
• Implemented algorithms in an open-source Julia package (Knockoffs.jl) with R and Python wrappers.

Main Results:

• Demonstrated effectiveness of proposed group knockoff methods through simulations.
• Applied methods to analyze albuminuria data from the UK Biobank.
• Showcased computational savings using correlation matrix approximations.

Conclusions:

• The expanded group knockoff framework provides a more powerful and precise approach for GWAS.
• The new methods and software facilitate the analysis of complex genetic data with correlated variables.
• This work enhances the ability to identify relevant genetic variants in large-scale association studies.