Lipid Class Prediction from MS1 Data using Gaussian Graphical Models

Area of Science:

• Biochemistry
• Analytical Chemistry
• Bioinformatics

Background:

• Untargeted liquid chromatography-mass spectrometry (LC-MS) enables comprehensive lipid profiling but is limited by unannotated features.
• Lipid class assignment offers a higher-level overview, complementing detailed structural analysis and supporting biological interpretation.
• Existing methods for lipid class prediction often rely on MS2 data, leaving a substantial portion of MS1-only features uncharacterized.

Purpose of the Study:

• To develop a systematic workflow for predicting lipid classes from MS1-only data in untargeted LC-MS.
• To improve the interpretation of complex lipidomics datasets by assigning classes to a larger number of features.
• To validate the utility of MS1-based lipid class prediction in biological studies, such as Alzheimer's disease research.

Main Methods:

• Development of GgmLipidClassifier (GLC), a method combining accurate-mass database searching with Gaussian graphical models (GGMs).
• Utilizing GGM-derived network structures from feature intensities to predict lipid class structure.
• Applying GLC to human serum and plasma datasets for lipid class prediction based on the LIPID MAPS Structure Database (LMSD) ontology.

Main Results:

• GLC achieved high accuracies: 82-90% at the main lipid class level and 72-86% at the subclass level across three human serum/plasma datasets.
• GLC demonstrated improved accuracy and reduced uncertainty compared to traditional closest-m/z matching.
• In an Alzheimer's disease study, GLC predictions revealed biologically plausible associations, including cholesterol, vitamin D3 derivatives, and plasmalogen glycerophosphoethanolamines, extending beyond existing annotations.

Conclusions:

• GLC provides robust lipid class predictions from MS1-only LC-MS data, significantly increasing the number of characterized lipid features.
• The method complements conventional analysis, enabling broader system-level interpretation of lipidomics data.
• GLC facilitates the discovery of novel lipid-disease associations by identifying features missed by traditional annotation methods.