RRmix
Overview
RRmix is a latent-factor mixture model for correcting batch effects in LC-MS metabolomics data, particularly when no internal standards or known controls are available. Instead of explicitly modeling known batch covariates, RRmix treats unobserved technical variation as latent factors: for all metabolites simultaneously it fits a hierarchical model with a random main effect + random compound-specific error variance + mixture structure.
The model decomposes observed metabolite intensities into biological signal, latent batch (or other technical) variation, and noise. It uses an EM algorithm to estimate parameters and, via a mixture indicator on the per-metabolite effects, identifies which metabolites are significantly associated with the biological condition (e.g., treatment vs control).
Because RRmix does not rely on internal standards or pre-specified technical covariates, it offers a flexible, data-driven approach to batch effect correction and differential abundance testing in high-dimensional metabolomics experiments.