Tests
A novel AI system designs proteins that efficiently bind to target molecules, which could advance drug design, disease understanding and more.
Every biological process in the body, from cell growth to immune responses, depends on interactions between molecules called proteins. Like a key in a lock, one protein can bind to another, helping to regulate critical cellular processes. Protein structure prediction tools like AlphaFold have already given us enormous insight into how proteins interact to perform their functions, but these tools cannot create novel proteins to directly manipulate these interactions.
Scientists can, however, create novel proteins that bind efficiently to target molecules. These binders can facilitate researchers accelerate progress in a wide range of research, including drug discovery, imaging cells and tissues, understanding and diagnosing diseases—even crop resistance to pests. While the latest approaches to machine learning Great progress has been made in protein design, but the process is still laborious and requires extensive experimental testing.
Today we present AlfaProteoour first AI system for designing novel, highly strong protein bonds to serve as building blocks for biological and health research. This technology has the potential to accelerate our understanding of biological processes and aid in drug discovery, biosensor development, and more.
AlphaProteo can generate novel protein bonds for a variety of target proteins, including VEGF-Awhich is associated with cancer and diabetes complications. For the first time, an AI tool was able to design an effective protein linker for VEGF-A.
The AlphaProteo method also achieves higher experimental success rates and binding affinities 3- to 300-fold better than the best existing methods for the seven target proteins we tested.
Exploring the convoluted ways proteins connect to each other
It is hard to design binding proteins that can bind tightly to the target protein. Time-honored methods are time-consuming and require many rounds of intensive laboratory work. Once the binders are created, they undergo additional rounds of experimentation to optimize their binding affinity so that they bind tightly enough to be useful.
Trained on a huge amount of protein data from Protein Data Bank (PDB) and over 100 million predicted structures from AlphaFold, AlphaProteo has learned the myriad ways that molecules bind to each other. Given the structure of a target molecule and a set of preferred binding sites on that molecule, AlphaProteo generates a candidate protein that binds to the target at those sites.
Illustration of a predicted protein-binding structure interacting with a target protein. The predicted protein-binding structure generated by AlphaProteo, designed to bind to the target protein, is shown in blue. The target protein, specifically the SARS-CoV-2 spike receptor binding domain, is shown in yellow
Demonstration of success on crucial protein binding targets
To test AlphaProteo, we designed binders for a variety of target proteins, including two viral proteins involved in infection, BHRF1 AND SARS-CoV-2 spike protein receptor binding domain, SC2RBD, and five proteins involved in cancer, inflammation, and autoimmune diseases, IL-7Rɑ, PD-L1, TrkA, Il-17A AND VEGF-A.
Our system has highly competitive binding success rates and best-in-class binding strengths. For seven targets, AlphaProteo generated in silico candidate proteins that bound strongly to the intended proteins during experimental testing.
Grid illustration of the predicted structures of seven target proteins for which AlphaProteo generated successful binding. Examples of binding tested in the soggy lab are shown in blue, protein targets are shown in yellow, and the intended binding regions are shown in shadowy yellow.
For one specific target, a viral protein BHRF188% of our candidate molecules bound successfully when tested in Google DeepMind Wet LabBased on tested targets, AlphaProteo binders bind on average 10 times stronger than the best existing design methods.
For another purpose, TrkAour adhesives are even stronger than the best adhesives previously designed for this purpose that have passed through multiple rounds of experimental optimization.
Bar graph showing the experimental in vitro success rates of AlphaProteo results for each of the seven target proteins compared to other design methods. Higher success rates indicate that fewer designs need to be tested to find effective binders.
Bar graph showing the best affinity for AlphaProteo designs without experimental optimization for each of the seven target proteins compared to the other design methods. Lower affinity indicates that the binding protein binds more tightly to the target protein. Note the logarithmic scale of the vertical axis.
We confirm our results
Apart from in silicon In order to validate and test AlphaProteo in our soggy lab, we engaged Pyotr Cherepanov, Katie Bentley AND David LV Bauer research groups from Francis Crick Institute to verify our protein bindings. In various experiments, they delved into some of our stronger SC2RBD and VEGF-A bindings. The research groups confirmed that the binding interactions of these bindings were indeed similar to those predicted by AlphaProteo. In addition, the groups confirmed that the bindings had a useful biological function. For example, some of our SC2RBD bindings were shown to prevent SARS-CoV-2 and some of its variants from infecting cells.
AlphaProteo’s performance suggests that it can drastically reduce the time needed for preliminary protein binding experiments for a wide range of applications. However, we know that our AI system has limitations, as it was unable to design productive bindings for the 8th target, TNFprotein associated with autoimmune diseases such as rheumatoid arthritis. We chose TNFɑ to challenge AlphaProteo because computational analysis showed that engineering the bonds would be extremely hard. We will continue to improve and expand the capabilities of AlphaProteo with the goal of ultimately addressing such challenging targets.
Achieving mighty binding is usually only the first step in designing proteins that can be useful in practical applications. There are many bioengineering hurdles to overcome in the research and development process.
Towards Responsible Development of Protein Design
Protein design is a rapidly evolving technology that has enormous potential to advance science in fields ranging from understanding what causes disease, to accelerating the development of diagnostic tests for viral epidemics, supporting more sustainable manufacturing processes, and even removing pollutants from the environment.
To address potential biosecurity risks, based on our long-standing approach to responsibility and safety, We are working with leading external experts to inform our phased approach to sharing this work and supporting community efforts to develop best practices, including NTI’s (Nuclear Threat Initiative) novel AI Bio Forum.
In the future, we will work with the scientific community to apply AlphaProteo to high-impact biology problems and understand its limitations. We have also explored its applications in drug design at Isomorphic Labs and are excited about what the future holds.
At the same time, we are continuously working to improve the success rate and affinity of AlphaProteo algorithms, expanding the range of design problems it can solve, and collaborating with scientists in machine learning, structural biology, biochemistry, and other disciplines to develop a responsible and more comprehensive protein design offering for the community.