Machine-Learning-Guided Peptide Drug Discovery: Development of GLP-1 Receptor Agonists with Improved Drug Properties

Area of Science:

• Pharmacology
• Biotechnology
• Computational Chemistry

Background:

• Peptide therapeutics are increasingly vital for treating metabolic diseases like diabetes and obesity.
• Machine learning (ML) in quantitative structure-activity relationship (QSAR) studies shows promise but faces data limitations in peptide drug discovery.

Purpose of the Study:

• To develop and validate a peptide drug discovery platform, streaMLine, for creating novel peptide hormone analogs.
• To identify potent, selective, and long-acting glucagon-like peptide-1 receptor (GLP-1R) agonists using secretin as a backbone.

Main Methods:

• Developed the streaMLine platform for integrated design, synthesis, screening, and ML-driven analysis of large peptide libraries.
• Synthesized and screened 2688 peptides, applying ML-guided QSAR to optimize GLP-1R agonists.
• Conducted in vivo profiling of a lead candidate in diet-induced obese mice.

Main Results:

• Identified multiple stable and potent GLP-1R agonist candidates through ML-guided QSAR analysis.
• The lead candidate, GUB021794, demonstrated potent body weight loss in obese mice.
• GUB021794 exhibited a pharmacokinetic profile suitable for once-weekly subcutaneous dosing.

Conclusions:

• The streaMLine platform effectively facilitates the discovery of peptide-based therapeutics.
• Optimized GLP-1R agonists with improved properties can be generated using secretin scaffolds and ML approaches.
• GUB021794 represents a promising candidate for further development in obesity treatment.