Deconvolution of composite mass spectra by artificial neural networks

In this paper we present the performance of an artificial neural network based model for deconvolution of composite mass spectra representing sixteen mixtures containing from, two to five components. Twenty-five electron impact mass spectra were used to train the neural network prior to examination of the mixtures. More than half of these compounds were either isomeric or isobaric with another compound present in the training set. The artificial neural network based model correctly identified individual components from these composite mass spectra even though all of the network training involved spectra from individual compounds. For mixtures of three or less compounds no false positive or false negative identifications were observed. When mixtures included up to five compounds a false positive identification of the same compound was observed in two of the mixtures. However, this false positive correlated with the presence of a chlorine isotope pattern rather than the presence of an isomeric or isobaric component.