Automated or semiautomated pattern recognition in multidimenan easy expansion to more general and multivariate cases. sional NMR spectroscopy is strongly hampered by the large number Recently, we reported a Bayesian method coupled to a multiof noise and artifact peaks occurring under practical conditions. A variate linear discriminant analysis (8). The actual implegeneral Bayesian method which is able to assign probabilities that mentation was limited to the use of local properties of resoobserved peaks are members of given signal classes (e.g., the class nance and artifact peaks. An important global property is of true resonance peaks or the class of noise and artifact peaks) was the spectral symmetry, which occurs in many types of homoproposed previously. The discriminative power of this approach is nuclear two-dimensional spectra such as NOESY and dependent on the choice of the properties characterizing the peaks.
TOCSY spectra. In the past this spectral feature was mainly
The automated class recognition is improved by the addition of a used for an improvement of the spectral quality by the symnonlocal feature, the similarities of peak shapes in symmetry-related metry enhancement (6,(9)(10)(11)(12)(13)(14). In the present paper we will positions. It turns out that this additional property strongly decreases the overlap of the multivariate probability distributions for true sig-show how it can be used for an efficient signal class recogninals and noise and hence largely increases the discrimination of true tion in two-dimensional NMR spectra.
resonance peaks from noise and artifacts.