Feature Selection by Lorentzian Peak Reconstruction for ^1NMR Post-Processing

In recent years, nuclear magnetic resonance spectroscopy (NMR) has become more and more popular in the field of metabolomic analysis. Analyzing and interpreting the obtained data is thus still challenging due to its complex and nontrivial characteristics. Further analysis of the obtained data is still mainly based on manual assignment, manual analysis and expert knowledge, and therefore time consuming. Common approaches towards automated post processing methods are often based on binning, which leads to loss of information in any case. This paper addresses an approach for reconstructing a one-dimensional NMR spectrum into a set of distinct lorentzian peak lines as an impressive feature selection and data reduction method and evaluates the performance on a real-world as well as on different simulated spectra.