DeepMind, Google's AI lab, has launched AlphaFold 2, a modern version of its AI model that can generate predictions for nearly every molecule in the Protein Data Bank, the largest database of freely accessible biomolecules worldwide. around the world.
Nearly five years ago, Google's AI Lab debuted AlphaFold, an AI system that can accurately predict the structure of multiple proteins in the human body. Since then, the laboratory has been constantly improving the system.
Isomorphic Laboratories is developing therapeutic products using the new co-developed AlphaFold model, which can describe different types of molecular structures important for treating diseases.
New AlphaFold model exceeds protein predictions. According to Google, the model can also accurately predict ligands (molecules that bind to receptor proteins and cause changes in the way cells communicate) and nucleic acids (which contain important genetic information), as well as subsequent processing. Modifications. Translation (chemical changes that occur after protein formation).
Google has found that predicting the protein structure of one molecule bound to another can be a useful tool in drug development. Because it can help scientists identify and develop new molecules that could become drugs.
Pharmaceutical researchers currently use computer simulations called “docking methods” to determine how proteins and molecules interact with each other. “Docking methods” require identification of the structure of the reference protein and proposed sites within that structure where the bound molecule might bind to another molecule.
Google explains that the modern version of AlphaFold does not require the use of reference protein structures or suggested positions, as the model can predict proteins that have not been previously described from a structural perspective, while the model simulates the way proteins and nucleic acids interact with other molecules. model; Google points out that this is not possible with current Installation Methods.
“Early analysis also shows that the current model significantly outperforms previous models on some protein structure prediction problems relevant to drug development, such as antibody binding,” Google wrote. “This tremendous advance in model performance demonstrates advances in the field of artificial intelligence.” The potential for scientific understanding of molecular machines. “The human body consists mainly of molecular machines.