An experimental validation of a Bayesian model for quantification in NMR spectroscopy.

The traditional peak integration method for quantitative analysis in nuclear magnetic resonance (NMR) spectroscopy is inherently limited by its ability to resolve overlapping peaks and is susceptible to noise. The alternative model-based approaches not only extend quantification capabilities to these challenging examples but also provide a means for automation of the entire process of NMR data analysis. In this paper, we present a general model for an NMR signal that, in a principled way, takes into account the effects of chemical shifts, relaxation, lineshape imperfections, phasing, and baseline distortions. We test the model using both simulations and experiments, concentrating on simple spectra with well-resolved peaks where we expect conventional analysis to be effective. Our results of quantifying mixture compositions compare favorably with the established methods. At high SNR (>40dB), all approaches usually achieve for these test systems an absolute accuracy of at least 0.01mol/mol for the concentrations of all species. Our model-based approach is successful even for SNR<20dB; it achieves 0.05-0.1mol/mol accuracy in cases where precise phasing is practically impossible due to high levels of noise in the data.