Classification of Mass Spectral Data Using Neural Networks

Abstract

A neural network has been trained to achieve the classification of mass spectral data. In order to keep the neural network small the input data for the network are generated from mass spectra by calculating spectral features. The neural network consists of two layers with only one neuron in the output layer and is trained to function as a binary classifier by using error back propagation. Three networks have been trained to be selective for steroids, barbituric acid derivatives and polycyclic aromatic hydrocarbons respectively. The neural networks showed good selectivity giving correct results with more than 90 % of the tested spectra.