Improving the Accuracy of Model-based Quantitative NMR

Abstract. We proposed an effective and computationally simple mechanism to improve the accuracy of model-based quantification in NMR data analysis. The proposed adjustment procedure aims to account for all useful signal left in the residual after the usual least squares fit, which can signify a case of model misspecification – a problem notoriously difficult to avoid in most model-based qNMR methods. Our alternative optimization criterion explicitly relies on the denoising of residual and smoothing the remaining baseline and is particularly effective in correcting errors in spectrum phasing. The results of analysis of experimental datasets obtained with high and medium field spectrometers indicate the accuracy improvement by 20–40 % compared to the usual least-squares model fit.