Adaptive weighted sum tests via LASSO method in multi-locus family-based association analysis

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Family-based multi-locus tests enhance genetic association studies by integrating information from multiple loci.
• Existing weighted sum methods have limitations, being applicable only to common or rare variants, leading to power loss, especially for rare variants.

Purpose of the Study:

• To propose a novel data-driven weight for multi-locus tests to improve statistical power for both common and rare variants.
• To introduce a LASSO-based approach for adaptive marker selection and weight construction in genetic association studies.

Main Methods:

• Utilized L1 regularization in Least Absolute Shrinkage and Selection Operator (LASSO) regression to develop a data-driven weight.
• Employed simulations for a dichotomous phenotype to evaluate the proposed method against existing multi-locus approaches.
• Applied the method to a real rheumatoid arthritis dataset (GAW15 Problem 2).

Main Results:

• The LASSO-based approach demonstrated superior statistical power compared to existing multi-locus methods while maintaining controlled type I error rates.
• The method successfully identified two allele groups with modest, non-significant individual SNP effects in the rheumatoid arthritis dataset, which traditional methods missed.
• Achieved high statistical significance (P < 0.00001) for detected allele groups.

Conclusions:

• The novel LASSO-based weight-choosing strategy offers a superior approach for multi-locus tests in genetic association studies.
• This method effectively improves power and handles both common and rare variants, addressing limitations of previous techniques.
• Demonstrated practical utility in identifying significant genetic associations in real-world datasets like rheumatoid arthritis.