A preview of the next generation of AlphaFold

Update in progress: Our latest AlphaFold model shows significantly improved accuracy and extends coverage beyond proteins to other biological molecules, including ligands.

Since its release in 2020, AlphaFold has revolutionized the way proteins and their interactions are understood. Google DeepMind and Isomorphic laboratories worked together to lay the foundation for a more powerful AI model that extends coverage beyond just proteins to the full range of biologically relevant molecules.

Today we are share an update on progress towards the next generation of AlphaFold. Our latest model can now generate predictions for almost any molecule in the Protein Data Bank (PDB), frequently achieving atomic precision.

It opens the door to new insights and significantly improves the accuracy of several key classes of biomolecules, including ligands (small molecules), proteins, nucleic acids (DNA and RNA), and those containing post-translational modifications (PTMs). . These different types and complexes of structures are essential to understanding biological mechanisms within the cell and are difficult to predict with high precision.

The model’s expanded capabilities and performance can help accelerate biomedical breakthroughs and realize the next era of “digital biology” — providing new insights into how disease transmission pathways work, genomics, biorenewable materials, plant immunity, potential therapeutic targets, drug design mechanisms, and new platforms enabling protein engineering and synthetic biology.

Beyond protein folding

AlphaFold constituted a fundamental advance for the prediction of single-chain proteins. AlphaFold-Multimere then expanded to complexes with multiple protein chains, followed by AlphaFold2.3, which improved performance and extended coverage to larger complexes.

In 2022, AlphaFold’s structure forecasts for nearly all cataloged proteins known to science were made available free of charge via the AlphaFold Protein Structure Databasein partnership with the EMBL European Bioinformatics Institute (EMBL-EBI).

To date, 1.4 million users in more than 190 countries have accessed the AlphaFold database, and scientists around the world have used AlphaFold’s predictions to advance research in every field, since acceleration of new malaria vaccines and move forward cancer drug discovery develop plastic-eating enzymes to fight against pollution.

Here we show AlphaFold’s remarkable abilities to predict precise structures beyond protein folding, generating highly accurate structure predictions for ligands, proteins, nucleic acids, and post-translational modifications.

Accelerate drug discovery

Initial analyzes also show that our model significantly outperforms AlphaFold2.3 on some protein structure prediction problems relevant to drug discovery, such as antibody binding. Additionally, accurately predicting protein-ligand structures is an extremely valuable tool for drug discovery, as it can help scientists identify and design new molecules that could become drugs.

The current industry standard is to use “docking methods” to determine interactions between ligands and proteins. These docking methods require a rigid reference protein structure and a suggested position at which the ligand should bind.

Our latest model sets a new bar for protein-ligand structure prediction by outperforming the best reported docking methods, without requiring a reference protein structure or ligand pocket location, enabling protein prediction completely new ones that have not been structurally characterized before.

It can also jointly model the positions of all atoms, allowing it to represent the full inherent flexibility of proteins and nucleic acids as they interact with other molecules – something not possible with conventional methods. mooring.

Here, for example, are three recently published therapeutically relevant cases where the structures predicted by our latest model (shown in color) closely match the experimentally determined structures (shown in gray):

1. PORCN: Anticancer molecule at the clinical stage linked to its target, together with another protein.
2. KRAS: Ternary complex with a covalent ligand (a molecular glue) of an important cancer target.
3. PI5P4Kγ: Selective allosteric inhibitor of a lipid kinase, with multiple pathological implications, including cancer and immunological disorders.

Isomorphic Labs is applying this next-generation AlphaFold model to therapeutic drug design, helping to quickly and accurately characterize many types of macromolecular structures important for disease treatment.

New understanding of biology

By opening up the modeling of protein and ligand structures as well as nucleic acids and those containing post-translational modifications, our model provides a faster and more accurate tool for examining fundamental biology.

An example concerns the structure of CasLambda bound to crRNA and DNApart of the CRISPR family. CasLambda shares the genome editing capability of CRISPR-Cas9 system, commonly known as “genetic scissors,” which researchers can use to modify the DNA of animals, plants, and microorganisms. The smaller size of CasLambda could allow for more efficient use in genome editing.

The latest release of AlphaFold’s ability to model such complex systems shows us that AI can help us better understand these types of mechanisms and accelerate their use for therapeutic applications. Other examples are available in our progress update.

Advancing scientific exploration

The dramatic increase in our model’s performance shows the potential of AI to dramatically improve scientific understanding of the molecular machines that make up the human body – and the broader natural world.

AlphaFold has already catalyzed major scientific advances around the world. Now, the next generation of AlphaFold has the potential to help advance scientific exploration at digital speed.

Our dedicated teams at Google DeepMind and Isomorphic Labs have made great progress on this critical work and we look forward to sharing our continued progress.