Retention time prediction for dereplication of natural products (CxHyOz) in LC-MS metabolite profiling

The detection and early identification of natural products (NPs) for dereplication purposes require efficient, high-resolution methods for the profiling of crude natural extracts. This task is difficult because of the high number of NPs in these complex biological matrices and because of their very high chemical diversity. Metabolite profiling using ultra-high pressure liquid chromatography coupled to high-resolution mass spectrometry (UHPLC–HR-MS) is very efficient for the separation of complex mixtures and provides molecular formula information as a first step in dereplication. This structural information alone or even combined with chemotaxonomic information is often not sufficient for unambiguous metabolite identification. In this study, a representative set of 260 NPs containing C, H, and O atoms only was analysed in generic UHPLC–HR-MS profiling conditions. Two easy to use quantitative structure retention relationship (QSRR) models were built based on the measured retention time and on eight simple physicochemical parameters calculated from the structures. First, an original approach using several partial least square (PLS) regressions according to the phytochemical classes provided satisfactory results with an easy calculation. Secondly, a unique artificial neural network (ANN) model provided similar results on the whole set of NPs but required dedicated software. The retention prediction methods described in this study were found to improve the level of confidence of the identification of given analytes among putative isomeric structures. Its applicability was verified for the dereplication of NPs in model plant extracts.