Google DeepMind has announced a new type of AI model for medical research that can predict the structure and interactions of all molecules in life.
The so-called "AlphaFold 3" AI model goes beyond the calculation of proteins, explained DeepMind, a subsidiary of the Google holding company Alphabet, and provides particularly accurate predictions for the interactions of proteins with other biomolecules in human cells.
The London-based DeepMind subsidiary Isomorphic Labs, which specialises in the use of artificial intelligence in drug research, also played a key role in the development of the AI model.
DeepMind had previously developed AlphaFold 2, a program that can calculate the three-dimensional shape of proteins.
Researchers believe that biological processes in the body and diseases can be better understood if the shape and thus also the function of the proteins are known.
In 2020, DeepMind achieved a fundamental breakthrough in the prediction of protein structures. So far, millions of researchers worldwide have used AlphaFold 2 to make discoveries in areas such as malaria vaccines, cancer treatments and enzyme design.
In an article published in the journal Nature, DeepMind presented the AI system AlphaFold 3 as a model that could predict the structure and interaction of all molecules of life with unprecedented accuracy.
"For the interactions of proteins with other types of molecules, we see an improvement of at least 50% compared to existing prediction methods, and for some important categories of interactions, we have doubled the prediction accuracy."
Google DeepMind also announced what it called the "AlphaFold Server". This free tool gives non-commercial researchers access to the capabilities of AlphaFold 3 to generate models of biological structures.
It enables biologists to generate large and complex protein structures on the screen with just a few mouse clicks. Isomorphic Labs is already working with pharmaceutical companies to utilise the potential of "AlphaFold 3" for the development of drugs.
