1910 Unlocks Macrocyclic Peptide Drugs Using PEGASUS AI Model

1910 is on a mission to “make undruggable targets a thing of the past” by “turning every pharma company into an AI company,” says founder and CEO, Jen Asher, PhD. The Sam Altman–backed AI drug discovery biotech closed out 2025 by rebranding from “1910 Genetics” to simply “1910,” signaling the company’s broad commitment to multi-modality drug discovery.

“2025 has been about us reintroducing ourselves to the world. ‘1910’ speaks to how we are bringing frontier AI research to bear in a modality agnostic manner,” Asher told GEN Edge.

Recently in the Journal of Medicinal Chemistry, 1910 published PEGASUS, an AI model that learns the rules for designing cell-permeable macrocyclic peptides and enables access to traditionally difficult to hit intracellular targets. The study provides new promise for a modality that can combine the oral convenience of small molecules with the high specificity of large biologics.

Asher describes PEGASUS as a “versatile tool” that accelerates the design–make–test cycle by functioning in both predictive and generative modes. Model capabilities include triaging compounds for synthesis, supporting lead optimization campaigns, and designing new starting peptides with desired properties. Notably, the study reports the first published cyclic peptides with more than two polar or ionizable fragments to achieve in vitro cell membrane permeability.

PEGASUS is one of two recent “flagship” publications that showcase “completely different modalities sprinting out of the same platform.” Last November in the Journal of Chemical Information and Modeling, 1910 published CANDID-CNS, an AI model that expands oral drug opportunities for neurological therapeutics by predicting small molecule blood–brain barrier (BBB) penetration within Beyond-Rule-of-5 (bRo5) chemical space.

bRo5 molecules intentionally violate Lipinski’s Rule of 5 (Ro5) guidelines for favorable oral bioavailability to tackle challenging molecular targets, such as flat surfaces for protein-protein interactions and complex protein binding sites.

With only about two percent of small-molecule drugs able to cross the BBB, accurate penetration prediction can identify promising candidates that are more likely to succeed in the clinic. CANDID-CNS achieved an 83% success rate for predicting bRo5 small molecule brain penetration and distribution, outperforming a 64% success rate for the industry standard, Pfizer’s CNS Multiparameter Optimization (CNS-MPO) score.

Asher says both models are “truly multimodal” and anchored by 1910’s unique wet lab biological data generation capabilities, which fuel the company’s internal pipeline and pharmaceutical partnerships to be announced in the future.

Founded in 2018, 1910 emerged from Y Combinator with a $4 million seed round led by OpenAI CEO, Sam Altman. Since that time, the Boston-based AI drug developer has landed a five-year commercial agreement with Microsoft, which offers 1910’s platform to biotechnology, government, and research institutions via three partnership models: co-discovery, co-engineering, and Platform-as-a-Service (PaaS).

The company name references the year that the first patient was diagnosed with sickle cell disease in the U.S., marking the first condition for which the field identified a molecular basis.

“When addressing disease, we want to go after targets that play a highly causative role in biology as opposed to an accessory role,” emphasized Asher. “1910 is our target selection North Star.”

Naturally permeable 

Macrocyclic peptides are stepping out of the margins of drug discovery and hitting the clinic. Last November, Merck reported that its macrocyclic peptide candidate for hypercholesterolemia, enlicitide, achieved statistically significant and clinically meaningful reductions in LDL cholesterol in a Phase III readout. If approved, the drug could become the first oral PCSK9 inhibitor, potentially disrupting a market dominated by injectable therapies.

Yet, Asher cautions that enlicitide’s uptake is contingent upon the administration of a permeability enhancer, which functions by damaging the cell membrane in the intestine, leading to clinically challenging effects, such as increased absorption of non-drug molecules and patient-to-patient drug variability.

“These issues, in addition to the higher cost of formulation, make normal physiological cell permeability the generally preferred method to achieve oral bioavailability,” she told GEN Edge.  

Achieving macrocyclic peptide permeability has remained elusive, as the traits that favor membrane passage—notably low polarity, high lipophilicity, and low molecular weight—frequently clash with therapeutically important features, including high potency and solubility. Solving this multiparameter optimization problem requires expanding upon existing wet lab ground truth biological datasets, which remain concentrated in high lipophilicity chemical space.

To generalize to new therapeutic areas, PEGASUS is trained on a unique multi-modal dataset from 1910’s proprietary Permeability Proxy Assay (1910 PPA), which generates billions of cyclic peptides separated by permeability-related characteristics, and solvent-dependent computational simulations based on quantum and molecular mechanics.

These data streams complement cell permeability data obtained from Caco-2 and Madin–Darby canine kidney (MDCK), in vitro systems that serve as the industry standard. On their own, data from Caco-2 and MDCK are inherently too low throughput for AI model training, time-consuming and expensive to obtain, and not optimized for cyclic peptides.

Asher says the work demonstrates how integrating wet lab and dry lab approaches can overcome major data barriers in AI-driven drug discovery. She adds that removing any one of the three data streams would diminish PEGASUS’s predictive and generative power, calling 1910’s surrogate assays a “breakthrough for AI model training.”

The field watches as these “not too big, not too small” drugs take one step closer toward clinical impact.