Google’s DeepMind AI lab has revealed AlphaFold 2, the latest iteration of an artificial intelligence model capable of generating predictions for nearly all molecules in the protein database, a vast open-access repository of biological molecules worldwide.
Approximately five years ago, Google’s AI lab first introduced the AlphaFold model, an AI system with remarkable accuracy in predicting the structures of various proteins within the human body. Since then, the lab has continued to refine the system.
Isomorphic Labs are now using the new AlphaFold model to design therapeutic drugs, aiding in characterizing different molecular structures crucial for disease treatment.
The capabilities of this new AlphaFold model go beyond protein prediction. Google claims it can also accurately predict the structures of ligands (molecules that bind to future proteins, causing changes in cell interactions), nucleic acids (molecules containing essential genetic information), and post-translation modifications (chemical changes that occur after protein formation).
Google points out that predicting the structure of a protein molecule related to another molecule can be a valuable tool in drug discovery. It can help scientists identify new molecules that could become drugs and design them. Currently, pharmaceutical researchers use computer simulation techniques known as “docking methods” to determine how proteins interact with other molecules, requiring the identification of a reference protein structure and the proposed binding site for the molecule associated with another.
Google explained that there is no need to use a reference protein structure or a proposed binding site with the latest version of AlphaFold.
The model can predict the structural aspects of previously uncharacterized proteins while simultaneously simulating how proteins and nucleic acids interact with other molecules, a level of modeling that is not achievable with current “docking methods.”
Google wrote: “Early analysis also shows that the current model significantly outperforms the previous model in some protein structure prediction challenges related to drug discovery, such as antibody binding. The remarkable leap in performance achieved by the model greatly enhances the scientific understanding of the molecular machines that make up the human body.”
