A novel spectra similarity measure

A new rnethod is described for calculating the similarity degree of two spectra. Its performance is optimized with similar, computer-generated H-1 NMR spectra. The method is compared with a recently proposed local cross-correlation method. Using a test set, its power to discriminate between related and unrelated H-1 NMR spectra is better than with the cross-correlation method. Better results are also obtained when comparing measured spectra of a database with the corresponding estimated ones or with estimated spectra of randomly assigned structures. Although, so far, it has only been tested with H-1 NMR spectra, due to the generality of the approach, the novel procedure can be applied to comparing other spectra or patterns as well. (c) 2005 Elsevier B.V. All rights reserved.