Machine Learning-Based Bioactivity Classification of Natural Products Using LC-MS/MS Metabolomics.

Area of Science:

• Natural Products Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Compound rediscovery is a major challenge in natural products drug discovery, hindering the exploration of novel compounds and wasting resources.
• Efficiently dereplicating known bioactive compounds is crucial for accelerating the discovery of new therapeutics.

Purpose of the Study:

• To develop and validate a novel machine learning (ML) framework for characterizing natural product bioactivity.
• To address the challenge of dereplicating previously discovered bioactive compounds in natural product drug discovery.

Main Methods:

• Leveraging liquid chromatography tandem mass spectrometry (LC-MS/MS) and untargeted metabolomics analysis.
• Utilizing the SIRIUS 5 metabolomics software suite for in-silico generated fragmentation spectra and molecular fingerprints (MFPs).
• Training an ML model on MFPs to predict compound drug class based on pharmacophores.

Main Results:

• The ML model achieved high accuracy (>93%) in classifying 21 diverse bioactive drug classes using experimental spectra.
• The framework enables rapid identification of bioactive scaffolds from LC-MS/MS data, even without reference spectra.
• Demonstrated the potential of ML combined with MFPs for dereplicating bioactive natural products based on pharmacophore.

Conclusions:

• The developed ML framework effectively characterizes natural product bioactivity and accelerates drug discovery.
• This approach streamlines the identification of novel bioactive scaffolds, expediting the isolation of potential antibacterial and antifungal agents.
• Highlights the power of integrating ML with metabolomics and computational tools for natural product research.