Identification of 1H-NMR spectra of N-linked oligosaccharides using artificial neural networks

A standard multi-layered perceptron is employed to implement the task of identification of a group of oligosaccharides (N-linked oligosaccharides) based on their 1D nuclear magnetic resonance (¹ H-NMR) spectra. The artificial neural network (ANN) was trained with 92 spectra representing 23 different compounds and tested with 23 other spectra (each compound represented once). This network achieved a classification rate of 100% during the training and testing phase while Bayesian classification applied to cluster distance measurements, produced a performance of less than 50%. A further advantage of the ANN identification routine is its insensitivity to instrument dependent variations of the spectra as long as the spectra are acquired at the correct field strength. This network is implemented on the world wide web for use by scientists around the world (http://www.ccrc.uga.edu)