Phenotypic Mapping of Metabolic Profiles Using Self-Organizing Maps of High-Dimensional Mass Spectrometry Data

A metabolic
system is composed of inherently interconnected metabolic
precursors, intermediates, and products. The analysis of untargeted
metabolomics data has conventionally been performed through the use
of comparative statistics or multivariate statistical analysis-based
approaches; however, each falls short in representing the related
nature of metabolic perturbations. Herein, we describe a complementary
method for the analysis of large metabolite inventories using a data-driven
approach based upon a self-organizing map algorithm. This workflow
allows for the unsupervised clustering, and subsequent prioritization
of, correlated features through Gestalt comparisons of metabolic heat
maps. We describe this methodology in detail, including a comparison
to conventional metabolomics approaches, and demonstrate the application
of this method to the analysis of the metabolic repercussions of prolonged
cocaine exposure in rat sera profiles.