A set of mass spectral classifiers has been developed to recognize
presence or absence of 70 substructures
or more general structural properties in a molecule.
Classification is based on numerical transformation
of
low resolution mass spectral data, automatic selection of appropriate
features, multivariate discriminant
methods, and estimation of the reliability of the classification
answer. Examples demonstrate applications
in structure elucidation together with automatic isomer generation as
well as combination with results obtained
by the CHEMICS system.