Deriving Lipid Classification Based on Molecular Formulas.

Area of Science:

• Metabolomics
• Lipidomics
• Bioinformatics

Background:

• Untargeted metabolomics faces challenges in accurate metabolite identification.
• Current methods often require orthogonal data (e.g., tandem MS, chromatography).
• Lipid category classification is crucial for biochemical interpretation in lipidomics.

Purpose of the Study:

• To develop a robust method for classifying elemental molecular formulas into lipid categories.
• To apply this classification to assignments generated by the SMIRFE algorithm.
• To facilitate biochemical interpretation in untargeted lipidomics without orthogonal data.

Main Methods:

• Utilized a Random Forest machine learning approach.
• Trained the model to predict lipid category and class from SMIRFE non-adducted molecular formula assignments.
• Evaluated classification performance using theoretical, data-derived, and artifactual molecular formulas.

Main Results:

• Achieved high average predictive accuracy (>90%) and precision (>83%) across eight LIPIDMAPS lipid categories.
• Successfully classified non-adducted molecular formulas generated by SMIRFE.
• Demonstrated utility for cross-spectrum assignment validation.

Conclusions:

• The developed method enables lipid classification of SMIRFE assignments without orthogonal information.
• Facilitates biochemical interpretation in untargeted lipidomics experiments.
• While insufficient for single-spectrum validation, it aids in cross-spectrum assignment validation.