Liquid-chromatography retention order prediction for metabolite identification

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Liquid Chromatography-tandem Mass Spectrometry (LC-MS/MS) is crucial for metabolite identification.
• Machine learning is transforming MS/MS data analysis, but LC data's potential is underutilized.
• Existing retention time prediction methods struggle with data heterogeneity.

Purpose of the Study:

• To develop a machine learning method for predicting the elution order of molecules in LC.
• To demonstrate that retention order is more conserved across different LC instruments than retention time.
• To enable training on heterogeneous LC data without extensive pre-processing.

Main Methods:

• A novel machine learning approach to predict molecular retention order in LC.
• Utilizing retention time data from diverse LC systems for training.
• Combining retention order predictions with MS/MS spectral data for enhanced metabolite identification.

Main Results:

• Retention order is significantly more conserved across instruments than absolute retention time.
• The method effectively learns molecular retention behavior from heterogeneous LC data.
• Combining retention order prediction with MS/MS scores improves metabolite identification accuracy.

Conclusions:

• Retention order prediction offers a robust strategy for metabolite identification using LC-MS/MS data.
• This method enhances the utility of diverse LC datasets, increasing available training data.
• The developed approach facilitates more accurate and efficient small molecule identification.