Sarah J. Nelson

A method for automatic quantification of one-dimensional spectra with low signal-to-noise ratio

Abstract A technique for processing noisy 1-D spectra is described. Unlike standard filters designed merely for noise suppression, this technique both smooths and automatically quantifies the peaks in a spectrum. There are three stages involved; identification of peaks from slowly varying background and random noise, estimation of peak parameters, and finally smoothing of the peaks on an individual basis. Simulated spectra, representative of those obtained from biological samples, are used to test the performance of the technique. The results agree very well with theoretical predictions. Finally, a comparison is made with the performance of two filters used for noise suppression, the matched filter and the maximum entropy method. The technique has considerable advantages over these two methods for the type of spectra considered here, the most important in practical terms being the ability to obtain automatic estimates of peak parameters together with a clear definition of the accuracy of these estimates.

The accuracy of quantification from 1D NMR spectra using the PIQABLE algorithm

Abstract The ability to produce accurate estimates of peak parameters in 1 D NMR spectra is of critical importance in interpreting experimental results, particularly for the analysis of in vivo spectra, where low signal to noise is common. The accuracy of the quantification obtainable using the automatic algorithm PIQABLE is reported here. The original version of PIQABLE produces reliable estimates of the areas of isolated peaks in low signal-to-noise spectra with a variable baseline. The algorithm has been extended to treat partially overlapping peaks and automatically estimate constant and linear phase corrections. A variety of simulated spectra has been analyzed in order to address four different topics: the accuracy of area estimates as a function of peak signal-to-noise ratio, the influence of variable baseline on area estimates, the effect of partially overlapping peaks, and the performance of the automatic phasing routines. The results underline the limitations imposed on any quantification method by the magnitude of random noise in the spectrum and the importance of employing statistical techniques to identify peaks and predict the accuracy of parameter estimates.