Google’s AI subsidiary, DeepMind, has announced the AlphaFold 3 model, which has the capability to predict protein structures, nucleic acids, and all details of living organisms. This model is expected to facilitate the work of researchers in medicine, agriculture, drug development, and more.
Previous versions of the model were only able to predict protein structures, but the new AlphaFold 3 model can now also predict DNA and RNA structures, as well as small biological molecules known as ligands, increasing its effectiveness in scientific research fields.
DeepMind says the new model demonstrates a 50% improvement in prediction accuracy compared to previous versions of the model.
Demis Hassabis, the CEO of DeepMind, stated during the company’s press conference: “The AlphaFold 3 model is a significant step in applying artificial intelligence to understand biological sciences and create models for them,” emphasizing the importance of the previous model in the field of biological research.
Regarding the operation of the AlphaFold 3 model, it includes a library of molecular structures, where researchers select the molecules they want to integrate, then the model uses the Diffusion mechanism to create a three-dimensional model of the new structure, which is the same mechanism used in some artificial intelligence systems like Stable Diffusion for image production.
DeepMind is working on implementing its advanced model in commercial applications and has announced a partnership with Isomorphic Labs pharmaceutical company to use the model in improving disease understanding and drug development. The company also provides researchers access to the model for non-commercial academic purposes.
Google states that it is collaborating with experts in the scientific community and political leaders to responsibly deploy the model, and has worked with specialists to understand the risks associated with the AlphaFold 3 model before its launch.
Although the model opens up immense possibilities in scientific research, Google emphasizes that some biosecurity experts fear that AI models may make it easier to design disease-causing agents and toxins with the use of other technologies, leading to their increased spread and harmful impact.
