Automated analysis of two-dimensional NMR spectra of mixtures by pattern recognition

Abstract

An improved procedure for pattern recognition in two-dimensional nuclear magnetic resonance spectra is described. The algorithm uses only the signs of the signals, and begins by searching for responses that are characteristic of three-spin subsystems, which are then identified as constituent parts of larger N-spin systems. The method relies on the comparison of cross-peak patterns obtained in two complementary 2D spectra. For the sake of demonstration, a conventional 1 D spectrum of a mixture of small molecules is decomposed into subspectra of the constituents. The chemical shifts and coupling constants resulting from the analysis are presented in a form which makes the presentation of the experimental 2D spectra superfluous.