Biohub, a network of nonprofit biomedical research institutes, has unveiled a new generation of Evolutionary Scale Models (ESM), an artificial intelligence system designed to predict, map, and generate proteins at large scale. The release reflects growing momentum in the use of AI for biological research and drug discovery, with the organization positioning the platform as an open framework for accelerating molecular science.
The system combines three main components: ESMC, a protein language model trained on roughly 2.8 billion protein sequences; ESMFold2, a structure prediction engine capable of modeling protein interactions and designing new proteins; and ESM Atlas, a database containing 6.8 billion protein sequences and more than 1 billion predicted structures. Together, the tools are intended to model the underlying biological principles that determine how proteins fold and function.
Proteins are essential to nearly all cellular processes, and their biological role depends on their three-dimensional structure. Traditional methods for protein design and testing often require years of laboratory work and extensive screening processes. Biohub’s approach aims to move much of that work into computational systems, enabling researchers to evaluate protein candidates digitally before conducting experiments in the lab.
The company highlighted progress in therapeutic protein design, particularly in developing proteins capable of binding to disease-related molecular targets. According to Biohub, ESMFold2 successfully generated high-affinity protein binders against five targets linked to cancer and immune disorders, including PD-L1, CTLA-4, EGFR, and PDGFRβ. Laboratory testing showed that several AI-designed proteins demonstrated strong binding activity and biological functionality.
One of the reported advances involves speed and scale. Biohub stated that the system can generate and rank tens of thousands of candidate proteins within days, significantly reducing the early-stage discovery timeline. Researchers also found that increasing computational resources improved the quality and success rate of designed binders, suggesting that larger compute capacity can directly improve experimental outcomes.
Biohub said ESMFold2 also achieved strong results in protein structure prediction benchmarks, particularly in antibody-antigen modeling, a key area for therapeutic development. Unlike many traditional protein prediction systems that rely heavily on evolutionary alignment techniques, ESMFold2 learns directly from massive biological datasets, allowing it to infer structural relationships from sequence information alone.
The release also includes ESM Atlas, which organizes billions of proteins into a searchable map intended to help researchers uncover hidden biological relationships and identify previously unknown functions. Biohub stated that the tools are being released under an open-source MIT license, with partnerships aimed at making the models broadly accessible to researchers and biotechnology developers.
The announcement highlights the expanding role of AI in life sciences, particularly as researchers seek faster and more scalable approaches to understanding biology and developing new therapies. By combining large-scale data, predictive modeling, and protein design, Biohub’s platform represents another step toward integrating artificial intelligence into core biomedical research workflows.
The post Biohub’s New Evolutionary Scale Models Advance AI-Driven Protein Engineering For Cancer And Immune Research appeared first on Metaverse Post.
