Enhancing lipid identification in LC-HRMS data through machine learning-based retention time prediction.

Area of Science:

• Biochemistry
• Computational Biology
• Analytical Chemistry

Background:

• Untargeted lipidomics using LC-MS/MS faces challenges in comprehensive peak identification.
• Accurate lipid annotation is crucial for understanding disease mechanisms, biomarker discovery, and drug screening.
• Machine learning-based retention time prediction can enhance lipid identification confidence.

Purpose of the Study:

• To develop and validate a machine learning model for predicting lipid retention times in LC-MS/MS untargeted lipidomics.
• To improve the accuracy and reliability of lipid peak annotation.
• To assess the impact of different molecular descriptors on model performance.

Main Methods:

• Development of a machine learning model utilizing molecular descriptors to predict retention times.
• Training and testing the model on LC-MS/MS data from untargeted lipidomics experiments.
• Comparison of molecular descriptors versus molecular fingerprints using Random Forest (RF) algorithms.
• External validation of the model's performance.

Main Results:

• The developed model achieved high correlation coefficients (0.998 training, 0.990 test) and low mean absolute errors (0.107 min training, 0.240 min test).
• External validation demonstrated strong performance with correlations of 0.991 and 0.978.
• Molecular descriptors outperformed molecular fingerprints when using the Random Forest algorithm.
• The model showed robust performance across different chromatographic systems.

Conclusions:

• The machine learning model significantly enhances lipid annotation accuracy in untargeted lipidomics.
• The model reduces errors in lipid identification, improving data analysis.
• This approach offers a reliable tool for various applications, including biomarker discovery and drug screening.