Principal Balances of Compositional Data for Regression and Classification using Partial Least Squares

TL;DR

This paper introduces a novel PLS-based method for constructing principal balances in high-dimensional compositional data, enhancing interpretability and maximizing response variability for regression and classification tasks.

Contribution

It develops a new PLS approach that constructs multiple orthonormal log-contrast balances, improving interpretability and performance over traditional methods.

Findings

Effective in capturing variability in compositional data

Improves interpretability in regression and classification

Performs well on simulated and real datasets

Abstract

High-dimensional compositional data are commonplace in the modern omics sciences amongst others. Analysis of compositional data requires a proper choice of orthonormal coordinate representation as their relative nature is not compatible with the direct use of standard statistical methods. Principal balances, a specific class of log-ratio coordinates, are well suited to this context since they are constructed in such a way that the first few coordinates capture most of the variability in the original data. Focusing on regression and classification problems in high dimensions, we propose a novel Partial Least Squares (PLS) based procedure to construct principal balances that maximize explained variability of the response variable and notably facilitates interpretability when compared to the ordinary PLS formulation. The proposed PLS principal balance approach can be understood as a generalized version of common logcontrast models, since multiple orthonormal (instead of one) logcontrasts are estimated simultaneously. We demonstrate the performance of the method using both simulated and real data sets.