AlphaFold goes beyond protein folding and will drive the next era of "digital biology"?

AlphaFold, which revolutionized life sciences, has been updated again——

Accuracy is significantly improved, and coverage is expanded from proteins to other biomolecules, including ligands.

Since its release in 2020, AlphaFold has revolutionized the way we understand proteins and their interactions, marking an important milestone in protein research.

Today, Google DeepMind stated in its official blog that through the joint efforts of them and the Isomorphic Laboratory, AlphaFold has been updated to the next generation, which will lay a more solid foundation for the application of AI in biological sciences.

It is reported that the new generation AlphaFold model can predict almost all molecules in the Protein Data Bank (PDB), and its prediction accuracy can reach the atomic level.

It not only opens up a new understanding of several key biomacromolecule classes, but also significantly improves prediction accuracy. These biomacromolecule classes include ligands (small molecules), proteins, nucleic acids (DNA and RNA), and biomacromolecules with post-translational modifications (PTMs), which are structural types and complexes that are critical to the understanding of biological mechanisms within cells.

Google DeepMind said that the extended functions and performance of this model will help accelerate breakthroughs in the biomedical field and promote humanity into the next era of "digital biology". It provides new insights and platforms for functional research on disease pathways, genomics, biorenewable materials, plant immunity, potential therapeutic targets, drug design mechanisms, protein engineering and synthetic biology.

AlphaFold has made a fundamental breakthrough in the prediction of single-chain proteins. AlphaFold-Multimer has been extended to complexes of multiple protein chains. AlphaFold2.3 not only improves the performance, but also expands the coverage to larger complexes.

In 2022, AlphaFold partnered with the European Bioinformatics Institute (EMBL-EBI) to make AlphaFold's structural predictions for almost all catalogued proteins known to the scientific community available for free through the AlphaFold Protein Structure Database.

To date, 1.4 million users from more than 190 countries have accessed the AlphaFold database, and scientists around the world have used AlphaFold's predictions to help advance research in various areas, from accelerating the development of new malaria vaccines, promoting the discovery of cancer drugs to developing plastic-eating enzymes to solve pollution problems.

In this study, Google DeepMind demonstrated AlphaFold's extraordinary ability in predicting precise structures beyond protein folding. It can make highly accurate structural predictions for ligands, proteins, nucleic acids, and post-translational modifications.

Figure | Performance of protein-ligand complexes, proteins, nucleic acids, and covalent modifications

In addition, the application of AlphaFold has also broadened the field of drug discovery.

The latest model significantly outperforms AlphaFold2.3 and industry standards on protein structure problems relevant to drug discovery, with particular attention paid to its performance in antibody binding prediction.

Traditional methods use rigid protein structures and docking methods to predict protein-ligand structures. However, the new generation AlphaFold model does not require this prior information, but shows higher accuracy and redefines the standards for predicting protein-ligand structures, allowing proteins with previously unknown structures to be predicted.

In addition, the model has the ability to jointly model the positions of all atoms, which can more fully reveal the flexibility of proteins and nucleic acids when interacting with other molecules.

In addition, in the latest recently published therapeutic cases, the structures predicted by the model are very close to the structures determined in the case experiments, including the binding of anti-cancer molecules (PORCN), covalent ligand binding of key cancer targets (KRAS), and the structure prediction of lipid kinase allosteric inhibitors (PI5P4Kγ).

Figure | Structural predictions of PORCN, KRAS, and PI5P4Kγ. The new AlphaFold model predictions are in color, and the case experiment predictions are in gray.

It is reported that Isomorphic Labs is applying the new generation AlphaFold model to therapeutic drug design to help quickly and accurately describe various types of macromolecular structures that are very important for treating diseases.

In addition, after unlocking the simulation of proteins, ligands, nucleic acids and post-translational modification structures, the model can provide a faster and more accurate tool for basic biological research.

For example, in the structure where CasLambda is bound to crRNA and DNA, CasLambda shares the gene editing capabilities of the CRISPR-Cas9 system, often referred to as “gene scissors,” which researchers can use to alter the DNA of plants, animals, and microorganisms, and the smaller size of CasLambda may make it more efficient in gene editing.

Figure | Predicted structure of CasLambda binding to crRNA and DNA

The ability of the new generation AlphaFold to model such complex systems suggests that AI can help us better understand these types of mechanisms and accelerate their application in therapeutics.

As the blog post states, “Google’s next-generation AlphaFold model brings unlimited potential to the scientific field and will provide deeper scientific understanding in the wider natural world. This huge progress heralds the bright prospects of AI in life sciences and provides strong support for future scientific exploration.”

Reference Links:

https://deepmind.google/discover/blog/a-glimpse-of-the-next-generation-of-alphafold/

https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/a-glimpse-of-the-next-generation-of-alphafold/alphafold_latest_oct2023.pdf

Author: Yan Yimi

Editor: Academic