Genome-wide regression and prediction with the BGLR statistical package

Area of Science:

• Genomics
• Statistical Genetics
• Computational Biology

Background:

• Genomic data analysis often involves regressions where parameter count (p) exceeds sample size (n), posing statistical and computational hurdles.
• Traditional methods struggle with these large-p-with-small-n scenarios, necessitating advanced statistical frameworks.

Purpose of the Study:

• To introduce the BGLR R-package for implementing Bayesian regression models tailored for large-p-with-small-n data.
• To showcase the package's capability in integrating diverse parametric and nonparametric shrinkage and variable selection techniques.

Main Methods:

• Utilizes Bayesian methods, including parametric variable selection, shrinkage, and semiparametric procedures like Bayesian reproducing kernel Hilbert spaces (RKHS) regressions.
• Employs a Gibbs sampler with scalar updates, leveraging efficient C and Fortran routines for computational performance.
• Supports continuous (censored or uncensored) and categorical (binary or ordinal) response variables.

Main Results:

• The BGLR package provides a unified framework for complex regression problems in genomics and beyond.
• Demonstrates the practical application and utility of Bayesian regression for large-scale data analysis.
• Highlights the flexibility of the implemented methods for various data types and analytical goals.

Conclusions:

• The BGLR R-package offers a powerful and versatile tool for addressing high-dimensional regression challenges in genomic and nongenomic research.
• Bayesian approaches provide a robust solution for integrating complex modeling strategies, enhancing data analysis capabilities.
• The package facilitates efficient and consistent implementation of advanced statistical methods for diverse scientific inquiries.