Accelerating natural product discovery with linked MS-genomics and language/transformer-based models

Area of Science:

• Microbiology
• Natural Product Chemistry
• Bioinformatics
• Artificial Intelligence

Background:

• Natural product discovery is crucial for new therapeutics.
• Characterizing microbial strain libraries is time-consuming.
• Integrating chemical and biological data can improve efficiency.

Purpose of the Study:

• To develop and validate an AI-driven framework for prioritizing microbial natural product producers.
• To leverage linked mass spectrometry (MS)-genome datasets for enhanced discovery.
• To accelerate the identification of high-potential natural product candidates.

Main Methods:

• Utilized language and transformer-based models for data analysis.
• Developed a framework for ranking microbial producers based on chem-bio data.
• Applied the framework to three case studies involving microbial strain libraries.

Main Results:

• The AI framework successfully prioritized microbial producers with 75-100% precision.
• Identified producers of diverse natural products across case studies.
• Demonstrated the ability to extract actionable insights from MS-genome data.

Conclusions:

• AI-enabled chem-bio characterization significantly accelerates natural product discovery.
• The approach enables access to microbial chemical diversity beyond existing knowledge.
• This method holds transformative potential for drug discovery and chemical biology.