ggmix

ggmix performs simultaneous SNP selection and population-structure-adjusted penalized linear mixed model analysis for high-dimensional genetic association and prediction.

Key Features:

• Simultaneous SNP Selection and Population Structure Adjustment: Integrates variable selection for single nucleotide polymorphisms (SNPs) with adjustment for population structure, enabling analysis when the number of fixed-effect predictors exceeds the sample size.
• Penalized Linear Mixed Model Framework: Implements a penalized linear mixed model with a single random effect to account for relatedness and confounding by population structure.
• Blockwise Coordinate Descent Algorithm: Uses a blockwise coordinate descent algorithm with automatic tuning-parameter selection, providing computational scalability and theoretical convergence guarantees.
• Enhanced Model Parsimony and Prediction Accuracy: Produces more parsimonious models and improved prediction accuracy compared to two-stage approaches or principal component adjustments, as demonstrated in simulations and real data examples.
• Robustness in Complex Genetic Scenarios: Maintains performance with highly correlated markers and when causal SNPs are included in the kinship matrix.

Scientific Applications:

• Polygenic Risk Scores Construction: Enables construction of polygenic risk scores by selecting multiple predictive SNPs to aid prediction of disease susceptibility.
• Mendelian Randomization Studies: Facilitates selection of instrumental variables for Mendelian randomization to support causal inference between genetic variants and traits or diseases.

Methodology:

Integrates SNP selection with population-structure adjustment in a single penalized linear mixed model and fits the model using a blockwise coordinate descent algorithm with automatic tuning-parameter selection and theoretical convergence guarantees.