Machine Learning-Assisted False Positive Detection in Metabolite Identification Workflows

Area of Science:

• Pharmacology and Cheminformatics
• Analytical Chemistry
• Computational Biology

Background:

• Metabolite identification is crucial for drug discovery and development.
• Liquid chromatography-mass spectrometry (LC-MS) data analysis is complex and prone to false positives.
• Accurate metabolite identification is essential for understanding drug metabolism and efficacy.

Purpose of the Study:

• To develop and validate a machine learning-based approach for improving the accuracy of false positive detection in metabolite identification.
• To enhance the efficiency and reliability of drug discovery workflows.
• To integrate expert knowledge with LC-MS data for robust metabolite characterization.

Main Methods:

• Developed a feature set for metabolite-related chromatographic peaks using expert knowledge.
• Integrated data from mass spectra, chromatographic signals, and kinetic profiles.
• Validated the approach using gradient boosting decision tree classifiers on diverse datasets (public and proprietary).

Main Results:

• Machine learning-assisted techniques significantly reduced false positive identifications.
• The developed feature set accurately characterized true and false positives.
• The method demonstrated high accuracy on various data sets, including small molecules and new modalities.

Conclusions:

• Machine learning significantly improves the accuracy of metabolite identification by reducing false positives.
• This approach enhances the efficiency and reliability of drug discovery and development pipelines.
• The integration of multi-modal data and expert knowledge is key to robust metabolite identification.