A metabolic profiling strategy for biomarker screening by GC-MS combined with multivariate resolution method and Monte Carlo PLS-DA

Abstract

A GC-MS based metabolic profiling method via a multivariate resolution method and Monte Carlo PLS-DA is proposed for screening potential biomarkers, and applied to Type 2 diabetes mellitus. The metabolic profiles of plasma samples from healthy control and Type 2 diabetes mellitus patient groups were obtained by GC-MS, and 25 compounds considered as endogenous metabolites excluding glucose were identified. With the help of a multivariate resolution method, qualitative and quantitative results of the metabolic profiles were extracted for subsequent multivariate statistical analysis. In order to select potential biomarkers, responsible for the classification of the two groups, Monte Carlo PLS-DA was introduced. The distribution of the regression coefficients of PLS-DA models corresponding to the metabolites was obtained. The levels of metabolites with all positive coefficients were considered as decreased from healthy controls to patients, and all negative coefficients were considered as increased. Univariate t-test was employed to check for metabolites whose levels changed significantly. Metabolites identified as potential biomarkers of Type 2 diabetes mellitus were ten in total, namely lactate, alanine, α-hydroxyisobutyric acid, phosphate, serine, pyroglutamic acid, palmitic acid, stearic acid, 1-monopalmitin and cholesterol. Finally, canonical correlation analysis was used to explore the correlation between the selected ten metabolites and blood glucose, which was considered to be a routine parameter reflecting the disease state. The results showed that the ten selected metabolites correlated well with blood glucose (r = 0.81, p = 0.03), and may be considered as possible biomarkers of Type 2 diabetes mellitus. The results demonstrated that the proposed method may be a useful tool to discover potential biomarkers of diseases.