A collaborative team from McMaster University, Duke University, and Cornell University has unveiled a breakthrough in drug discovery: an AI system that can design peptide drugs to target disease related proteins once thought undruggable.

Published August 13, 2025, in Nature Biotechnology, the study describes how the new tool, called PepMLM, was adapted from a language model originally built for human communication similar to those used in chatbots but retrained to understand the language of proteins.

Unlike traditional drug design tools that depend on knowing a protein’s 3D structure, PepMLM works from the protein’s amino acid sequence alone. This makes it possible to target a much wider range of proteins, including those without stable structures a common challenge in tackling cancers, neurodegenerative diseases, and viral infections.

Most drug design tools rely on knowing the 3D structure of a protein, but many critical disease targets don’t have stable structures,said Pranam Chatterjee, senior author of the study, who led the work at Duke and now serves on the faculty at the University of Pennsylvania. PepMLM changes the game by designing peptide binders using only sequence information.

In laboratory experiments, the AI successfully generated short amino acid chains peptides that bound to, and in some cases helped destroy, disease related proteins. Targets included proteins implicated in cancer, reproductive disorders, Huntington’s disease, and even live viral infections.

At McMaster University, Christina Peng, a PhD student in the Truant Lab, led the Huntington’s disease experiments. It’s exciting to see these AI designed peptides work inside cells to break down toxic proteins, Peng said. This could offer new hope for diseases where conventional drugs have failed.

Cornell University’s Matthew DeLisa and Hector Aguilar tested the AI generated peptides on viral proteins, while Chatterjee  Duke team developed the model and conducted initial validation. Ray Truant, professor in McMaster’s Department of Biochemistry & Biomedical Sciences, emphasized the platform versatility.We can now bind any protein to any other protein. That means we can degrade harmful ones, stabilize helpful ones, or precisely control protein modifications depending on the therapeutic goal.

The researchers are already advancing next generation tools, such as PepTune and MOG DFM, to enhance peptide stability, targeting, and delivery in the body. Our ultimate goal is a general purpose, programmable peptide therapeutic platform one that starts with a protein sequence and ends with a viable drug, Chatterjee said.

This work was supported by the CHDI Foundation, Wallace H. Coulter Foundation, The Hartwell Foundation, the National Institutes of Health, and the Krembil Foundation, among others. Chatterjee and first author Tianlai Chen are co inventors on U.S. patent applications related to PepMLM, and Chatterjee and co author DeLisa hold financial interests in UbiquiTx, Inc, a biotech company developing programmable protein based therapies.

Source: https://healthsci.mcmaster.ca/researchers-re-engineer-ai-language-model-to-target-previously-undruggable-disease-proteins/