Computer-assisted interpretation of carbon-13 nuclear magnetic resonance spectra applied to structure elucidation of natural products

Abstract

The reduction of chemical and spectroscopic data to their structural Implications Is a major component of the computer model of the structure elucidation process developed at Arizona State University. This paper compares five classifiers for binary 13C-NMR spectral data. While selection of the best technique for spectral Interpretation Is largely dependent upon the particular situation, several trends are evident from this investigation. If a thorough enough data set exists and an ample supply of computer time Is available, a search of the data set seeking the most similar member should prove valuable. When dealing with binary data, the Tanlmoto similarity measure correctly classifies unknown compounds more often than the conventional distance measure (Nearest Neighbor). Of the three nonsearching procedures Investigated, maximum likelihood and distance from the mean have comparable predictive abilities, and both procedures correctly classify spectra more frequently than the dot product classifier. Examples are presented which demonstrate that the manner In which questions are posed greatly affects predictive ability.

Structure evaluation using predicted carbon-13 spectra

“…Consequently, any of the observed triplets could be assigned to C (6). As a consequence of better substructural models in the data base (higher shell values, column 3), other CH2 resonances are associated with narrower predicted ranges [e.g., C(12), C (17)]. It would, in general, be appropriate for these resonances to be matched to the observed triplets first; subsequently, resonances with wider predicted ranges could be assigned.…”

Computer-Based Structure Determination: Then and Now

“…In the case of binary classifiers, performance is described in terms of the probability that a compound predicted to be in a class is actually in that class. Applications to 1 H and 13 C NMR, IR, and MS have been studied. − …”

“…Applications to 1 H and 13 C NMR, IR, and MS have been studied. [32][33][34] More recently, neural networks, a form of supervised pattern recognition, have been applied to spectrum interpretation. 35 As with other pattern recognition methods, the relationship between spectral properties and structural features in a molecule need not be specified in advance.…”

Rebuilding Connectivity Matrices from Two-Atom Fragments Using the Genetic Algorithm

“…8 Due to varying conditions and equipment, experimental and reference spectra were not always perfectly superimposable, and more sophisticated lookup methods had to be devised, using a variety of statistical techniques to retrieve the closest ("nearest neighbor") spectrum. 9 At the time the problems were compact storage and quick retrieval. There was no attempt at interpreting the spectra, nor would it have been feasible considering the then state of the art.…”