Modifying Chromatography Conditions for Improved Unknown Feature Identification in Untargeted Metabolomics

Area of Science:

• Systems Biology
• Analytical Chemistry
• Biochemistry

Background:

• Untargeted metabolomics is crucial for systems biology but struggles with identifying many detected features.
• Liquid chromatography-tandem mass spectrometry (LC-MS/MS) is used for identification, but poor spectra limit annotation.
• Many features lack sufficient spectral quality for successful compound identification.

Purpose of the Study:

• To improve compound identification performance in untargeted metabolomics using human plasma samples.
• To explore LC-MS/MS parameter optimization for enhanced spectral quality and metabolite annotation.
• To develop a machine learning model for classifying unidentified features.

Main Methods:

• Investigated alterations in gradient length, mass loading, and precursor ion exclusion for RPLC and HILIC.
• Performed manual review of spectral matches to establish thresholds for semi-automated MS/MS data analysis.
• Developed a localized machine learning model to classify remaining unidentified features.

Main Results:

• Optimized LC-MS/MS methods increased metabolite identification from 214 to 2052 unique compounds.
• 68.0% of newly identified features were detectable and quantifiable in a standard 20-min LC-MS run.
• Machine learning classified 576 (HILIC) and 749 (RPLC) unidentified features as high-priority targets.

Conclusions:

• A simple strategy significantly deepens untargeted metabolomics data annotation with minimal extra effort.
• Optimized LC-MS/MS parameters and machine learning improve metabolite identification rates and data utility.
• The study provides a practical approach for more comprehensive metabolomic profiling.