Google DeepMind has unveiled an AI system called Alpha Proteo that can design new, novel proteins that actually bind to target molecules-a potential revolution in drug design and research into diseases.Alpha Proteo could design new protein binders for a wide array of target proteins, including VEGF-A, which are implicated in cancer and complications of diabetes. The work represents a first in using an AI tool to design a protein binder against VEGF-A.
Most striking is the performance of this system, where experimental success rate and binding affinities for seven tested target proteins were up to 300-fold better than those from existing methods.
The model, which is trained on vast amounts of protein data in the Protein Data Bank and over 100 million predicted structures from AlphaFold, learned how to intricately interact at the binding surfaces of molecules. Given a target molecule structure and preferred binding locations, the system produces a candidate protein designed to bind at those specific sites.
To validate the capabilities of AlphaProteo, the team designed binders for a wide range of target proteins, including viral proteins involved in infection and those associated with cancer, inflammation, and autoimmune diseases. Most promisingly, high success rates of binding were found, with best-in-class binding strengths observed across the board.
For instance, where the target of interest was the viral protein BHRF1, 88% of AlphaProteo candidate molecules bound successfully in wet lab testing. On average, the AlphaProteo binders demonstrated binding 10 times stronger than that of the best existing design methods across targets tested.
Performance here indicates this system can reduce the time it takes to carry out initial experiments that involve binders of proteins in a huge array of applications. Simultaneously, the team recognizes some of the limitations of AlphaProteo-it is unable to design successful binders against TNFɑ, a protein involved in autoimmune diseases like rheumatoid arthritis.
For this reason, Google DeepMind works with a variety of external experts to inform its step-by-step approach to sharing this work and contributing to community efforts in developing best practices, including the NTI's new AI Bio Forum.The team will work with the scientific community to apply AlphaProteo on important biological problems to learn about limitations as the technology keeps improving.They are also working on applications in drug design at Isomorphic Labs.While AlphaProteo is a huge step toward protein design, strong binding is in general just the first step toward practical applications in protein design. There are still very many bioengineering challenges that need to be overcome along research and development.In any case, the breakthrough by Google DeepMind has great potential to accelerate progress in many research activities, such as in drug development, cell and tissue imaging, disease understanding and diagnosis, and even in the resistance of crops to pests.
