This article concerns a new expert system for elucidating structures of organic compounds from their spectra, where two principally different molecular generators are used. The system is based on a database of 13C NMR spectra (135,000 structures), a library of molecular fragments and their 13C NMR subspectra (more than 400,000) and a library of spectrum-to-structure correlations for both NMR and IR spectra. At the first stage, the system tries to automatically determine the structure of the unknown compound using fragments taken from the library of 13C NMR spectra. The structures are generated by joining fragments that have common atoms. Knowledge of the molecular mass and the molecular formula is not needed for this stage. In case of failure the program automatically determines the molecular formula from the molecular mass (different formulas can have the same mass) and then forms the sets of fragments. The structures are generated with the help of another generator designed for constructing structures from non-overlapping fragments. The libraries of spectrum-to-structure correlations are used for determination of the molecular formula as well as for spectral filtration of fragments and structures. The most preferable structure is selected by predicting 13C and 1H NMR spectra. Some examples of problem solution are cited.
Read full abstract