Simulation of mass spectra of noncyclic alkanes and alkenes using artificial neural network

A 37-10-44 artificial neural network (ANN) was successfully developed for t he simulation of mass spectra of noncyclic alkanes and alkenes. A total of 37 topological descriptors was selected using multiple linear regression (M LR) technique and was employed as inputs for the ANN. Forty-four outputs of the ANN represent the percent of total ion current (TIC%) in 44 positions. A collection of 117 noncyclic alkanes and 145 noncyclic alkenes was chosen as data set which were in the range of C5-10 H8-22. The data set was rando mly divided into a training set consisting of 236 molecules and a predictio n set consisting of 26 compounds. The results obtained indicate that except for the molecular ion peak, the predicted values of the m/z positions as w ell as their intensities are in good agreement with the experiment. Compari son of the SEC and SEP values of the ANN with those of the MLR models revea ls the superiority of the ANN over that of the regression model for simulat ion of the mass spectra of organic compounds. (C) 2000 Elsevier Science B.V . All rights reserved.