The rapid development of metabolomics has significantly advanced health and disease related research. However, metabolite identification remains a major analytical challenge for untargeted metabolomics. While the use of collision cross-section (CCS) values obtained in ion mobility-mass spectrometry (IM-MS) effectively increases identification confidence of metabolites, it is restricted by the limited number of available CCS values for metabolites. Here, we demonstrated the use of a machine-learning algorithm called support vector regression (SVR) to develop a prediction method that utilized 14 common molecular descriptors to predict CCS values for metabolites. In this work, we first experimentally measured CCS values (Ω) of ∼400 metabolites in nitrogen buffer gas and used these values as training data to optimize the prediction method. The high prediction precision of this method was externally validated using an independent set of metabolites with a median relative error (MRE) of ∼3%, better than conventional theoretical calculation. Using the SVR based prediction method, a large-scale predicted CCS database was generated for 35 203 metabolites in the Human Metabolome Database (HMDB). For each metabolite, five different ion adducts in positive and negative modes were predicted, accounting for 176 015 CCS values in total. Finally, improved metabolite identification accuracy was demonstrated using real biological samples. Conclusively, our results proved that the SVR based prediction method can accurately predict nitrogen CCS values (Ω) of metabolites from molecular descriptors and effectively improve identification accuracy and efficiency in untargeted metabolomics. The predicted CCS database, namely, MetCCS, is freely available on the Internet.

The use of collision cross-section (CCS) values derived from ion mobility-mass spectrometry (IM-MS) has been proven to facilitate lipid identifications. Its utility is restricted by the limited availability of CCS values. Recently, the machine-learning algorithm-based prediction (e.g., MetCCS) is reported to generate CCS values in a large-scale. However, the prediction precision is not sufficient to differentiate lipids due to their high structural similarities and subtle differences on CCS values. To address this challenge, we developed a new approach, namely, LipidCCS, to precisely predict lipid CCS values. In LipidCCS, a set of molecular descriptors were optimized using bioinformatic approaches to comprehensively describe the subtle structure differences for lipids. The use of optimized molecular descriptors together with a large set of standard CCS values for lipids (458 in total) to build the prediction model significantly improved the precision. The prediction precision of LipidCCS was externally validated with median relative errors (MRE) of ∼1% using independent data sets across different instruments (Agilent DTIM-MS and Waters TWIM-MS) and laboratories. We also demonstrated that the improved precision in the predicted LipidCCS database (15 646 lipids and 63 434 CCS values in total) could effectively reduce false-positive identifications of lipids. Common users can freely access our LipidCCS web server for the following: (1) the prediction of lipid CCS values directly from SMILES structure; (2) database search; and (3) lipid match and identification. We believe LipidCCS will be a valuable tool to support IM-MS-based lipidomics. The web server is freely available on the Internet ( http://www.metabolomics-shanghai.org/LipidCCS/ ).