Generalized adaptive intelligent binning of multiway data

Abstract

NMR metabolic fingerprinting methods almost exclusively rely upon the use of one-dimensional (1D) 1H NMR data to gain insights into chemical differences between two or more experimental classes. While 1D 1H NMR spectroscopy is a powerful, highly informative technique that can rapidly and nondestructively report details of complex metabolite mixtures, it suffers from significant signal overlap that hinders interpretation and quantification of individual analytes. Two-dimensional (2D) NMR methods that report heteronuclear connectivities can reduce spectral overlap, but their use in metabolic fingerprinting studies is limited. We describe a generalization of Adaptive Intelligent binning that enables its use on multidimensional datasets, allowing the direct use of nD NMR spectroscopic data in bilinear factorizations such as principal component analysis (PCA) and partial least squares (PLS).