Quantitative structure–retention relationships models for prediction of high performance liquid chromatography retention time of small molecules: Endogenous metabolites and banned compounds

Abstract

Quantitative structure–retention relationship (QSRR) is a technique capable of improving the identification of analytes by predicting their retention time on a liquid chromatography column (LC) and/or their properties. This approach is particularly useful when LC is coupled with a high-resolution mass spectrometry (HRMS) platform. The main aim of the present study was to develop and describe appropriate QSRR models that provide usable predictive capability, allowing false positive identification to be removed during the interpretation of metabolomics data, while additionally increasing confidence of experimental results in doping control area. For this purpose, a dataset consisting of 146 drugs, metabolites and banned compounds from World Anti-Doping Agency (WADA) lists, was used. A QSRR study was carried out separately on high quality retention data determined by reversed-phase (RP-LC–HRMS) and hydrophilic interaction chromatography (HILIC-LC–HRMS) systems, employing a single protocol for each system. Multiple linear regression (MLR) was applied to construct the linear QSRR models based on a variety of theoretical molecular descriptors. The regression equations included a set of three descriptors for each model: ALogP, BELe6, R2p and ALogP2, FDI, BLTA96, were used in the analysis of reversed-phase and HILIC column models, respectively. Statistically significant QSRR models (squared correlation coefficient for model fitting, R2=0.95 for RP and R2=0.84 for HILIC) indicate a strong correlation between retention time and the molecular descriptors. An evaluation of the best correlation models, performed by validation of each model using three tests (leave-one-out, leave-many-out, external tests), demonstrated the reliability of the models. This paper provides a practical and effective method for analytical chemists working with LC/HRMS platforms to improve predictive confidence of studies that seek to identify small molecules.