BOSTON — 1910 said it has published PEGASUS, a multimodal artificial intelligence model designed to predict and generate cell-permeable macrocyclic peptides, marking what the company described as a significant advance in peptide drug discovery.
The research was published as a Featured Article in the Journal of Medicinal Chemistry and reports the first known examples of macrocyclic peptides containing more than two polar or charged fragments that demonstrate in vitro cell permeability. According to the company, the work addresses a longstanding challenge in peptide design, where increasing polarity and charge has traditionally made it difficult for molecules to cross cell membranes.
Macrocyclic peptides are considered a promising class of therapeutics due to their potential for oral bioavailability and intracellular activity, but their development has been constrained by limited permeability data and a historical bias toward hydrophobic designs. Existing datasets have been small and unrepresentative, limiting the ability of AI models to generalize across broader chemical space.
PEGASUS was developed to overcome these limitations by integrating multiple data modalities, including high-throughput wet-lab proxy biological data generated through 1910’s permeability proxy assay, computational simulation data from solvent-dependent quantum mechanical models, and geometric and biological embeddings that capture structural features relevant to permeability. Together, these datasets enable the model to learn permeability-related characteristics across a wide range of macrocyclic peptide chemistries, including highly polar and charged regions that have traditionally been difficult to design.
The company said access to this expanded chemical space is critical, noting that restricting designs to low-polarity peptides not only reduces the number of viable sequences but may also increase the risk of off-target effects and toxicity.
“In drug discovery, AI has always been constrained by the lack of large, high-quality biological datasets,” said Jen Asher, Ph.D., founder and chief executive officer of 1910. “By generating billions of experimental data points and integrating them with physics-based simulations, we built a model that expands the therapeutic possibilities for macrocyclic peptides.”
In retrospective validation studies, the PEGASUS predictive framework improved hit rates by 13.1 percent when used as a pre-synthesis filter, outperforming existing deep-learning approaches. Its generative component produced 33 macrocyclic peptides with polarity and charge profiles similar to FDA-approved therapeutics. Of those synthesized and tested, four demonstrated permeability consistent with in vivo oral bioavailability, which the company said represents a first for peptides in this chemical regime.
“Cell permeability is essential for oral drug delivery, yet there remains limited chemical overlap between macrocyclic peptides that are routinely designed to be permeable and those that have achieved clinical success,” said Cole Baker, AI research scientist II at 1910 and lead author of the publication. “Our work helps bridge this gap to enable the design of orally bioavailable macrocyclic peptide therapeutics.”
According to 1910, PEGASUS was developed within its ITO platform and is designed to function both as a high-accuracy predictor of permeability and as a generative system capable of designing drug-like peptide candidates with improved solubility, polarity, and charge characteristics. The company said the publication provides a broader framework for applying multimodal data integration to advance new therapeutic modalities.
