Principal Component Analysis (PCA) Loading and Statistical Tests for Nuclear Magnetic Resonance (NMR) Metabolomics Involving Multiple Study Groups
Lin Jiang, Hunter Sullivan, Bo Wang
Abstract
Metabolomics is an interdisciplinary area that integrates knowledge of instrumentation, data science, and biochemistry. Metabolomics studies the changes in a large number of metabolites after various treatments using analytical platforms. However, the interpretation approaches have not been completely investigated. Principal component analysis (PCA) is an unsupervised method that describes high throughput metabolite data, which is different from supervised approaches such as partial least-squares discriminant analysis (PLS-DA) which frequently has overfitting problems. The interpretation of PCA loadings, especially for studies with multiple study groups, is not well developed for metabolomics. In this study, a new method was reported that integrates PCA loading values with the commonly used statistical t-test analysis to significantly improve the convenience and efficiency of interpretation. The method was demonstrated using practical studies of NMR metabolomics on the extracts from sea anemone that were treated with six atrazine concentrations. The results indicated that the approach is suitable for multiple groups of metabolomics for early-stage discoveries, such as low concentrations and potentially longitudinal studies. In summary, this methodology may be critical in studies such as environmental metabolomics with various stimuli factors where the data interpretation was previously incompletely developed.