Maximizing the quality of NMR automatic metabolite profiling by a machine learning based prediction of signal parameters

Abstract

The quality of automatic metabolite profiling in NMR datasets in complex matrices can be compromised by the multiple sources of variability in the samples. These sources cause uncertainty in the metabolite signal parameters and the presence of multiple low-intensity signals. Lineshape fitting approaches might produce suboptimal resolutions or distort the fitted signals to adapt them to the complex spectrum lineshape. As a result, tools tend to restrict their use to specific matrices and strict protocols to reduce this uncertainty. However, the analysis and modelling of the signal parameters collected during a first profiling iteration can further reduce the uncertainty by the generation of narrow and accurate predictions of the expected signal parameters. In this study, we show that, thanks to the predictions generated, better profiling quality indicators can be outputted and the performance of automatic profiling can be maximized. Thanks to the ability of our workflow to learn and model the sample properties, restrictions in the matrix or protocol and limitations of lineshape fitting approaches can be overcome.