LipidOA: A Machine-Learning and Prior-Knowledge-Based Tool for Structural Annotation of Glycerophospholipids

The Paternò–Büchi (PB) reaction is a carbon–carbon double bond (C═C)-specific derivatization reaction that can be used to pinpoint the location(s) of C═C(s) in unsaturated lipids and quantitate the location of isomers when coupled with tandem mass spectrometry (MS/MS). As the data of PB-MS/MS are increasingly generated, the establishment of a corresponding data analysis tool is highly needed. Herein, LipidOA, a machine-learning and prior-knowledge-based data analysis tool, is developed to analyze PB-MS/MS data generated by liquid chromatography–mass spectrometry workflows. LipidOA consists of four key functional modules to realize an annotation of glycerophospholipid (GPL) structures at the fatty acyl-specific C═C location level. These include (1) data preprocessing, (2) picking C═C diagnostic ions, (3) de novo annotation, and (4) result ranking. Importantly, in the result-ranking module, the reliability of structural annotation is sorted via the use of a machine learning classifier and comparison to the total fatty acid database generated from the same sample. LipidOA is trained and validated by four PB-MS/MS data sets acquired using different PB reagents on mass spectrometers of different resolutions and of different biological samples. Overall, LipidOA provides high precision (higher than 0.9) and a wide coverage for structural annotations of GPLs. These results demonstrate that LipidOA can be used as a robust and flexible tool for annotating PB-MS/MS data collected under different experimental conditions using different lipidomic workflows.

• LipidOA, a data analysis tool integrating machine learning and prior knowledge is developed, enabling structural annotation of glycerophospholipids (GPLs) at the C=C location level with using PB-MS/MS data.
• LipidOA allows the detection of novel lipid species which may not have been reported before, and also contains a ranking system that sorts many annotations in three tiers and effectively places high precision (>0.9) annotations in Tier 1.
• LipidOA can be applied to analyze PB-MS/MS data which are collected on different MS instruments or using different workflows.