Presentation Abstract

NMR spectroscopy is an essential tool for structural analysis of molecules. Yet the way from a set of NMR spectra to a verification of a presumed molecular formula or to an elucidation of an unknown is still extremely challenging, complex, and not always as successful as one might hope. Multidimensional techniques certainly help a lot but one should keep in mind that, in practice, the most widely used experiment is still the basic 1D NMR which, when properly interpreted, can provide a wealth of useful information with minimum acquisition time (cost) and sample quantity. This is particularly true for proton spectra of organic molecules.

However, a detailed analysis of 1D NMR spectra, especially an automatic one, is often hindered by insufficient resolution (real and digital), presence of extra lines (solvent and impurities), artifacts due to dead time and shimming, spectral lines overlap, strong coupling effects, etc.

Though the exact theory of spectral analysis has been worked out decades ago, automatic analysis of 1H-NMR spectra is generally still performed by algorithms based on simple first order rules, loosing a great amount of valuable information. This is due mostly to the fact that while spectral simulation (the direct problem) is relatively straightforward, deducing a structure from a spectrum (the inverse problem) may be excruciatingly complex and possibly lead to multiple solutions.

In this talk we present our current efforts aiming at an expert system which will overcome the above difficulties, take advantage of any a-priori knowledge supplied by the chemist, and get the most out of 1D NMR spectra.

The system comprises a number of auxiliary algorithms for boosting resolution, detecting spectral peaks, isolating multiplets, global spectral deconvolution, quantum-mechanical simulation of spin systems of any size, fitting of experimental spectra, novel ways to sort out the coupling structure between various multiplets, etc.

Many of these algorithms have an immediate, independent value of their own and can be extended in a natural way also to 2D spectra.