New AI Tools Promise a Revolution in Peptide-Based Drug Discovery
The future of drug development may lie in targeting the incredibly specific building blocks of proteins – peptides. Recent advancements in artificial intelligence are unlocking the potential of these small protein fragments, offering a “universal tool” for creating new therapies, according to researchers in the field.
For years, scientists have sought ways to identify adn utilize peptide binding sites within proteins. Now, a wave of sophisticated algorithms is accelerating this process.Researchers at the University of Toronto have been at the forefront, developing tools like PepNN-Struct and PepNN-Seq, which predict where peptides bind based on a protein’s structure or genetic code. These models feed into PepMLM, further refining the identification process.
Building on this foundation,the Toronto team recently unveiled PepFlow,a deep-learning model capable of predicting peptide structures themselves.This is crucial as peptides are inherently flexible molecules, existing in multiple shapes, and understanding these conformations is key to understanding their function and how to best utilize them therapeutically.
“Peptides are vital biological molecules and are naturally dynamic, so we need to model their different conformations to understand their function,” explains Philip M. Kim, PhD, professor of molecular genetics and computer science at the University of Toronto, and a leading figure in this research. Dr.kim, who holds a Canada Research Chair in machine learning in protein and peptide science, also co-developed the PepNN models.
The potential of peptide-based therapies is already evident in successful drugs like glucagon-like peptide-1 analogues, such as Ozempic, which are used to manage diabetes and obesity. Though, these new AI tools aim to expand the scope of treatable conditions dramatically.
While these deep-learning models are still evolving, their promise is driving innovation. Several biotechnology companies have already spun out from the PepMLM research, dedicated to refining these models and applying them to a wider range of diseases.
A key question for the future, Dr. kim notes,is whether these models can accurately predict peptide behavior in new situations,or if they simply excel at analyzing data they’ve already been trained on. Even if the models primarily “interpolate” from existing data, that capability alone could be incredibly valuable.
This research was supported by funding from Duke University, cornell University, the CHDI Foundation, the Wallace H. Coulter Foundation, the hartwell Foundation, the Krembil Foundation, and the National Institutes of health. Dr. kim disclosed financial ties to Fable Therapeutics, TBG Therapeutics, and Zymedi as a cofounder and consultant.
