Increasing Compound Identification Rates in Untargeted Lipidomics Research with Liquid Chromatography Drift Time–Ion Mobility Mass Spectrometry

Unknown
metabolites represent a bottleneck in untargeted metabolomics
research. Ion mobility–mass spectrometry (IM-MS) facilitates
lipid identification because it yields collision cross section (CCS)
information that is independent from mass or lipophilicity. To date,
only a few CCS values are publicly available for complex lipids such
as phosphatidylcholines, sphingomyelins, or triacylglycerides. This
scarcity of data limits the use of CCS values as an identification
parameter that is orthogonal to mass, MS/MS, or retention time. A
combination of lipid descriptors was used to train five different
machine learning algorithms for automatic lipid annotations, combining
accurate mass (m/z), retention time
(RT), CCS values, carbon number, and unsaturation level. Using a training
data set of 429 true positive lipid annotations from four lipid classes,
92.7% correct annotations overall were achieved using internal cross-validation.
The trained prediction model was applied to an unknown milk lipidomics
data set and allowed for class 3 level annotations of most features
detected in this application set according to Metabolomics Standards
Initiative (MSI) reporting guidelines.