Neural network methodologies for mass spectra recognition

Abstract The purpose of this work was to establish the methodology for automated mass spectra recognition using neural networks. Four different neural networks techniques were tested (backpropagation, improved Kohonen network, ART2 and multilayered perceptron) and compared on simulated mass spectra samples. The testing environment set for all four neural networks spectra recognition systems showed almost the same efficiency and robustness for improved Kohonen type network as well as for the ART2 system. According to test results, both systems can be recommended for practical use in mass spectra recognition. The stage of development of neural network methodologies is gaining on maturity. It is evident that their use is especially powerful in applications dealing with numerous parameters and their correlations (some of them unknown) such as mass spectrometry, Auger electron spectroscopy (AES), X-ray diffraction (XRD) etc.