Machine Learning to Develop Peptide Catalysts-Successes, Limitations, and Opportunities.

Area of Science:

• Catalysis
• Biochemistry
• Computational Chemistry

Background:

• Peptides are versatile modular catalysts for chemical transformations.
• Identifying peptides with specific catalytic activity is challenging due to the vast number of possible amino acid sequences.

Purpose of the Study:

• To develop and validate a machine learning (ML) workflow for optimizing peptide catalysts.
• To identify novel peptide catalysts for specific chemical reactions using ML.

Main Methods:

• Development of an ML workflow for peptide catalyst optimization.
• Creation of a universal training set (UTS) with 161 catalysts.
• In silico screening of a library of approximately 30,000 tripeptide members.

Main Results:

• The ML workflow successfully identified peptide catalysts for the conjugate addition of aldehydes to nitroolefins.
• A novel peptide catalyst was identified for a previously uncatalyzed stereoselective annulation reaction.
• Demonstrated the efficacy of a data-driven approach compared to expert-knowledge-guided optimization.

Conclusions:

• Machine learning provides an efficient strategy for discovering and optimizing peptide catalysts.
• The developed ML workflow and UTS can accelerate the identification of peptide catalysts for diverse chemical transformations.
• Data-driven optimization offers a powerful alternative to traditional methods in peptide catalyst design.