gpls

gpls performs generalized partial least squares classification to reduce dimensionality and classify biological samples from high-throughput data (e.g., gene expression) within a generalized linear model framework using iteratively reweighted partial least squares (IRWPLS) and Firth's bias-reduction procedure.

Key Features:

• Dimensionality Reduction: Extends partial least squares (PLS) to handle classification in high-dimensional settings with many covariates and few samples.
• Generalized Linear Model Framework: Frames classification within a generalized linear model (GLM) to model response distributions appropriate for categorical outcomes.
• Iteratively Reweighted Partial Least Squares (IRWPLS): Uses iterative reweighting to refine estimates and improve classification accuracy.
• Firth's Procedure: Incorporates Firth's bias-reduction method to address (quasi)separation and reduce bias in logistic regression parameter estimates.
• Comparative Performance: Reported to achieve lower classification error rates compared to two-stage PLS and other classifiers in benchmark studies.

Scientific Applications:

• Gene Expression Analysis: Classifying samples based on gene expression profiles into biological categories or disease states.
• Multi-group Classification: Handling classification tasks involving more than two groups in high-throughput biological data.

Methodology:

Extends partial least squares for classification within a generalized linear model using iteratively reweighted partial least squares (IRWPLS) and incorporates Firth's procedure to mitigate (quasi)separation in logistic regression.