The Compound Characteristics Comparison (CCC) approach: a tool for improving confidence in natural compound...

Area of Science:

• Metabolomics
• Computational Chemistry
• Machine Learning

Background:

• Metabolomics relies heavily on Liquid Chromatography - High Resolution Mass Spectrometry (LC-HRMS) for compound identification.
• A significant challenge in LC-HRMS is the high number of unidentified metabolites, requiring expert interpretation.
• Existing in silico fragmentation simulators often require manual refinement by chemists, especially for complex plant natural products.

Purpose of the Study:

• To develop a supervised machine learning approach for predicting molecular substructures from isotopic patterns.
• To enhance the accuracy and efficiency of compound identification in metabolomics.
• To assist expert chemists in annotating unknown metabolites, particularly in plants.

Main Methods:

• A supervised machine learning model, Compounds Characteristics Comparison (CCC), was trained on a database of grape metabolites.
• The CCC approach predicts molecular substructures based on isotopic patterns.
• The CCC predictions were integrated as scoring terms within Metfrag 2.2 for querying MS/MS spectra.

Main Results:

• The CCC approach demonstrated good accuracy in predicting molecular substructures.
• Integration of CCC with Metfrag improved the ranking of correct compound candidates.
• The method increased user confidence in selecting correct annotations from mass spectrometry data.

Conclusions:

• The Compounds Characteristics Comparison (CCC) approach effectively complements existing identification strategies like fragmentation simulators and formula calculators.
• CCC assists in the crucial task of compound identification in metabolomics.
• The CCC algorithm is available as an R package, adaptable for other biological matrices with additional training data.