In collaboration with the AI-powered drug discovery company Isomorphic Labs, Google DeepMind has released a new iteration of the protein structure prediction model, AlphaFold 2. The latest model can predict structures for most molecules in the Protein Data Bank (PDB) with high accuracy, often reaching atomic precision, in various important biomolecule categories, including small molecules (ligands), proteins, nucleic acids (DNA and RNA), and molecules with post-translational modifications (PTMs).

Considering that the comparison systems use known protein structures as a basis, it is a significant achievement that AlphaFold latest outperforms traditional systems like open-source molecular modeling simulation software AutoDock Vina in the accuracy of ligand docking, even when it starts with only protein sequences and ligand information. Secondly, it shows improvement over AlphaFold 2.3, particularly in predicting protein-protein structures, with a notable enhancement in antibody binding structures.

Thirdly, when it comes to protein-nucleic acid interfaces, AlphaFold latest excels compared to competing methods, and in RNA structure prediction, it outperforms automated techniques. However, it falls slightly short of the top CASP15 entrant, which involves manual expert intervention.

Lastly, AlphaFold extends its capabilities to predict the structure of additional components such as bonded ligands, glycosylation, and modified residues or nucleotides.

In 2021, Google DeepMind solved the 50-year-old grand protein folding challenge with AlphaFold. Since proteins form the building blocks of humans, the AI system soon became a boon for life sciences. AlphaFold was a fundamental breakthrough for single-chain protein prediction.

Even though OpenAI’s ChatGPT broke the internet when it was unveiled in 2022, the most-cited paper of the year, with 1331 citations, came from the European Molecular Biology Laboratory (EMBL-EBI) and DeepMind, focusing on the AlphaFold Protein Structure Database.

Similarly, the second most-cited paper, with 1138 citations, was from the Max Planck Institute for Multidisciplinary Sciences, and it centered on making protein folding accessible to all through ColabFold. The framework has been used extensively to find its real-life use cases, including the prediction of protein structures for the COVID-19 outbreak—SARS-CoV-2, drugs for liver cancer, gene therapy, and more. And now, with the latest model, it is going to accelerate the process. 

Read more: 2022 was The Year of Protein Folding Models. Wait, What?

The post Google DeepMind, Isomorphic Labs Unveil Next Generation of AlphaFold appeared first on Analytics India Magazine.

In collaboration with the AI-powered drug discovery company Isomorphic Labs, Google DeepMind has released a new iteration of the protein structure prediction model, AlphaFold 2. The latest model can predict structures for most molecules in the Protein Data Bank (PDB) with high accuracy, often reaching atomic precision, in various important biomolecule categories, including small molecules (ligands), proteins, nucleic acids (DNA and RNA), and molecules with post-translational modifications (PTMs).

Considering that the comparison systems use known protein structures as a basis, it is a significant achievement that AlphaFold latest outperforms traditional systems like open-source molecular modeling simulation software AutoDock Vina in the accuracy of ligand docking, even when it starts with only protein sequences and ligand information. Secondly, it shows improvement over AlphaFold 2.3, particularly in predicting protein-protein structures, with a notable enhancement in antibody binding structures.

Thirdly, when it comes to protein-nucleic acid interfaces, AlphaFold latest excels compared to competing methods, and in RNA structure prediction, it outperforms automated techniques. However, it falls slightly short of the top CASP15 entrant, which involves manual expert intervention.

Lastly, AlphaFold extends its capabilities to predict the structure of additional components such as bonded ligands, glycosylation, and modified residues or nucleotides.

In 2021, Google DeepMind solved the 50-year-old grand protein folding challenge with AlphaFold. Since proteins form the building blocks of humans, the AI system soon became a boon for life sciences. AlphaFold was a fundamental breakthrough for single-chain protein prediction.

Even though OpenAI’s ChatGPT broke the internet when it was unveiled in 2022, the most-cited paper of the year, with 1331 citations, came from the European Molecular Biology Laboratory (EMBL-EBI) and DeepMind, focusing on the AlphaFold Protein Structure Database.

Similarly, the second most-cited paper, with 1138 citations, was from the Max Planck Institute for Multidisciplinary Sciences, and it centered on making protein folding accessible to all through ColabFold. The framework has been used extensively to find its real-life use cases, including the prediction of protein structures for the COVID-19 outbreak—SARS-CoV-2, drugs for liver cancer, gene therapy, and more. And now, with the latest model, it is going to accelerate the process. 

Read more: 2022 was The Year of Protein Folding Models. Wait, What?

The post Google DeepMind, Isomorphic Labs Unveil Next Generation of AlphaFold appeared first on Analytics India Magazine.