Data sharing in PredRet for accurate prediction of retention time: Application to plant food bioactive compounds
Dorrain Yanwen Low, Pierre Micheau, Ville Koistinen, Kati Hanhineva, László Abrankó, Ana Rodriguez‐Mateos, Andreia Bento‐Silva, Christof Van Poucke, Conceição Almeida, Cristina Andrés‐Lacueva, Dilip K. Rai, Esra Çapanoğlu, Francisco A. Tómas‐Barberán, Fulvio Mattivi, G. Schmidt, Gözde Gürdeniz, Kateřina Valentová, Letizia Bresciani, Lucie Petrásková, Lars Ove Dragsted, Mark Philo, Marynka Ulaszewska, Pedro Mena, Raúl González‐Domínguez, Rocío Garcı́a-Villalba, Senem Kamiloğlu, Sonia de Pascual‐Teresa, Stéphanie Durand, Wiesław Wiczkowski, Maria Rosário Bronze, Jan Stanstrup, Claudine Manach
Abstract
Prediction of retention times (RTs) is increasingly considered in untargeted metabolomics to complement MS/MS matching for annotation of unidentified peaks. We tested the performance of PredRet (http://predret.org/) to predict RTs for plant food bioactive metabolites in a data sharing initiative containing entry sets of 29-103 compounds (totalling 467 compounds, >30 families) across 24 chromatographic systems (CSs). Between 27 and 667 predictions were obtained with a median prediction error of 0.03-0.76 min and interval width of 0.33-8.78 min. An external validation test of eight CSs showed high prediction accuracy. RT prediction was dependent on shape and type of LC gradient, and number of commonly measured compounds. Our study highlights PredRet's accuracy and ability to transpose RT data acquired from one CS to another CS. We recommend extensive RT data sharing in PredRet by the community interested in plant food bioactive metabolites to achieve a powerful community-driven open-access tool for metabolomics annotation.