Penalized Logistic Regression Analysis for Genetic Association Studies of Binary Phenotypes

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genetic association studies for binary phenotypes face challenges with data sparsity due to unbalanced case-control ratios or rare variants.
• Data sparsity can lead to biased maximum likelihood estimators (MLEs) and inflated type 1 errors in logistic regression.
• Penalized-likelihood methods are employed to mitigate sparse-data bias in genetic analyses.

Purpose of the Study:

• To develop and evaluate a penalized logistic regression method using log-F priors to address data sparsity in genetic association studies.
• To improve the accuracy of parameter estimation and reduce type 1 errors in the presence of sparse genetic data.
• To provide an easily implementable method for rare variant analysis in binary trait genetics.

Main Methods:

• A two-step approach is proposed: first, estimating the shrinkage parameter (m) using a marginal likelihood maximization (via MCEM or Laplace approximation).
• Second, applying log-F-penalized logistic regression with the estimated m for variant association analysis.
• The method utilizes data augmentation for easy implementation with standard statistical software.

Main Results:

• Simulation studies indicate the proposed log-F-penalized approach exhibits lower bias and mean squared error compared to existing shrinkage methods.
• The method demonstrated improved statistical properties in handling sparse genetic data.
• The approach was successfully illustrated on a real-world genetic association study dataset.

Conclusions:

• A novel method for single rare variant analysis in binary phenotypes using logistic regression penalized by log-F priors has been developed.
• The proposed method effectively addresses data sparsity and associated biases.
• The approach is readily extensible for controlling confounding factors like population structure and genetic relatedness via data augmentation.