Genomic-Enabled Prediction of Ordinal Data with Bayesian Logistic Ordinal Regression

Area of Science:

• Genomic Prediction
• Statistical Genetics
• Bioinformatics

Background:

• Genomic prediction models typically assume continuous, normally distributed response variables.
• Existing models for ordinal phenotypes, like the Bayesian probit ordinal regression (BPOR), are less frequently implemented in genomic prediction, especially when the sample size is smaller than the number of parameters (n << p).
• Bayesian logistic ordinal regression (BLOR) is rarely used in this context due to implementation challenges.

Purpose of the Study:

• To propose a novel Bayesian logistic ordinal regression (BLOR) model for genomic-enabled prediction of ordinal phenotypes.
• To develop an efficient computational method for the BLOR model using Pólya-Gamma data augmentation.
• To demonstrate the utility of the proposed BLOR model as an alternative to probit and logit link functions in genomic prediction.

Main Methods:

• Development of a BLOR model incorporating a Pólya-Gamma data augmentation approach.
• Implementation of a Gibbs sampler that yields full conditional distributions analogous to the BPOR model.
• Evaluation of the proposed BLOR model through simulation studies and analysis of two real-world datasets.

Main Results:

• The proposed BLOR model, utilizing Pólya-Gamma data augmentation, provides a computationally efficient Gibbs sampler.
• The BLOR model demonstrated comparable or superior performance to probit and logit link models in analyzing ordinal data.
• The BPOR model is shown to be a special case of the proposed BLOR model, highlighting its flexibility.

Conclusions:

• The proposed BLOR model is a viable and effective alternative for genomic-enabled prediction of ordinal traits.
• The Pólya-Gamma data augmentation approach simplifies the implementation of BLOR models in genomic prediction scenarios.
• This work expands the methodological toolkit for analyzing complex genetic data with categorical outcomes.