An artificial neural network for classifying cross peaks in two-dimensional NMR spectra

A simulated neural network is described that has been trained to classify cross peaks in the 2D NMR spectra of biological macromolecules. The trained network has then been used to classify previously unseen data. The network is able to distinguish between authentic cross peaks and spectral artifacts, such as those arising from presaturation of water, noise, and t₁ noise. Moreover, the network is able to recognize genuine peaks whose shapes have been modified, for example, by overlap with other real or spurious peaks. Herein, the training and performance of the network are demonstrated for a NOESY spectrum.