Detection of cross peaks in two-dimensional NMR by cluster analysis

Abstract

Considering the complexity of two-dimensional NMR spectra (l-3), it appears desirable to delegate at least part of their analysis to a computer. Several successful attempts have been made in this direction in the past (4-6), although a complete analysis by computer of a 2D spectrum for a complex biomolecule is not yet practical. Several approaches are conceivable for the analysis of 2D spectra: (i) recognition of basic patterns and logical analysis of the connectivity network (4, 5), (ii) simulation of partial or entire spectra and iterative fitting of experimental data, (iii) library search and pattern match. Many procedures require the initial recognition of cross peaks in an experimental 2D spectrum. A recently proposed method takes advantage of the basic antiphase square patterns that form building blocks in phase-sensitive 2D correlation spectroscopy (COSY) (4, 5). A computer search with a quadratic mask of variable dimension produces a reduced data set consisting of positional coordinates and “active” coupling constants (side length of basic square). In the reduced data set it is then possible to identify cross peaks by finding entries with neighboring coordinates and equal active coupling constants. In a sense, this procedure combines the search for cross peaks and their analysis into a single step. Obviously, a resolved and identifiable cross-peak multiplet structure is a prerequisite for a successful analysis of this type. Approaches that require a full analysis of the multiplet structure may lead to complications in the following situations: (i) The multiplet structure is only partially resolved and some comers of the antiphase squares are missing. (ii) Limited sensitivity permits the detection of partial multiplets only. (iii) Strong coupling effects complicate the multiplet structure and partially destroy the basic antiphase multiplet structure. (iv) Only absolute value spectra are available. (v) 2D NOE spectra must be analyzed which normally contain badly resolved in-phase squares as building blocks. It should be noted that none of these complications impedes a visual detection of cross peaks. It is therefore desirable to adopt some of the visual detection principles to the computer procedure in order to circumvent possible problems. The eye, in contrast to the computer, initially detects coarse features, it does a “cluster analysis.” Only at a later stage does it focus on the detailed fine structure whereby symmetry features are of much help. These are the two approaches followed in this communication.