Researchers at Google DeepMind have published a white paper on AlphaProteo, a new AI system that can design novel proteins to bind to specific molecular targets. This tech could accelerate research across biology and medicine, potentially leading to new therapeutics and diagnostic tools.
Protein binders are valuable tools in drug development and biotechnology. They have shown promise in targeting cancer cells, blocking viral infections, and modulating immune responses. However, traditional methods for developing protein binders are time-consuming, often requiring extensive laboratory work.
AlphaProteo aims to dramatically speed up this process. The AI was trained on vast protein datasets, including structures from the Protein Data Bank and millions of AlphaFold predictions. This allowed it to learn the intricate ways molecules bind to each other.
In tests, AlphaProteo designed successful binders for seven diverse target proteins, including two involved in viral infections and five linked to cancer, inflammation, and autoimmune diseases. For one viral target called BHRF1, 88% of AlphaProteo's designs bound successfully when tested in the lab.
Notably, AlphaProteo became the first AI method to design effective binders for VEGF-A, a protein associated with cancer and diabetic retinopathy. On average, AlphaProteo's binders showed 3 to 300 times stronger binding affinity than the best existing methods across the targets tested.
The Francis Crick Institute helped validate the results, confirming that some of AlphaProteo's binders could prevent SARS-CoV-2 variants from infecting cells.
While promising, AlphaProteo does have limitations. It failed to design binders for a challenging autoimmune disease target called TNF𝛼. Google DeepMind says it will continue improving the system's capabilities.
The company is taking a cautious approach to sharing this technology, working with biosecurity experts to develop best practices. They plan to collaborate with the scientific community to apply AlphaProteo to important biological problems while expanding its capabilities.
This breakthrough in AI-powered protein design could significantly accelerate the early stages of drug discovery and biological research. However, extensive work remains to translate these initial results into real-world applications.