EM Adaptive LASSO-A Multilocus Modeling Strategy for Detecting SNPs Associated with Zero-inflated Count Phenotypes

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Count data with excess zeros are common in genetic association studies, particularly in rheumatology.
• Standard count distributions (Poisson, ZIP, NB, ZINB) may not adequately handle excessive zeros or model misspecification.
• Limited research exists on the performance of these methods in genetic association studies, especially concerning variable selection.

Purpose of the Study:

• To investigate the performance of existing statistical methods for zero-inflated count data in genetic association studies.
• To introduce and evaluate a novel penalized regression approach with an adaptive LASSO penalty for zero-inflated count phenotypes.
• To assess the methods' robustness to model misspecification and their utility in variable selection.

Main Methods:

• Simulated genetic association data under various disease models and linkage disequilibrium patterns.
• Compared existing state-of-the-art methods with a novel penalized regression approach incorporating data-adaptive weights.
• Utilized a nested EM algorithm with coordinate descent for parameter estimation and simultaneous variable selection.

Main Results:

• The proposed penalized regression method exhibited optimal performance in prediction accuracy and empirical power, particularly with increasing sample sizes.
• The novel method demonstrated robustness in the presence of multicollinearity.
• Competing methods showed uncontrollable Type I error rates when models were misspecified.

Conclusions:

• The novel penalized regression approach offers a flexible and robust solution for analyzing zero-inflated count data in genetic association studies.
• This method effectively handles multicollinearity and model misspecification, outperforming existing approaches.
• The findings highlight the importance of appropriate statistical modeling for accurate genetic association analysis with complex count phenotypes.