Evaluation of ensemble Monte Carlo variable selection for identification of metabolite markers on NMR data

Abstract

The aim of this study was to investigate the potential of the recently developed ensemble Monte Carlo Variable Selection (EMCVS) method to identify the relevant portions of high resolution 1H NMR spectra as a metabolite fingerprinting tool and compare to a widely used method (Variable importance on projection (VIP)) and recently proposed variable selected methods i.e. selectivity ratio (SR) and significance multivariate correlation (sMC). As case studies two quantitative publicly available datasets: wine samples, urine samples of rats, and an experiment on mushroom (Agaricus bisporus) were examined. EMCVS outperformed the three other variable selection methods in most cases, selecting fewer chemical shifts and leading to improved classification of mushrooms and prediction of onion by-products intake and wine components. These fewer chemical shift regions facilitate the interpretation of the NMR spectra, fingerprinting and identification of metabolite markers.