In the struggle for leadership in the field of artificial intelligence, Google made a big breakthrough, presenting the next update of its Alphafold system to predict the 3D structure of proteins. The new version has received advanced opportunities and is now capable of modeling not only proteins, but also other most important biological molecules, including DNA.
According to Demis Hassabis, General Director of DeepMind, AlphaFold 3 is exactly the tool that is extremely necessary to open new drugs. “You need to see how a small molecule contacts the drug how tightly it joins it and what other molecules interacts with,” he explained in an interview Wired on the eve of the publication of the scientific article about Alphafold 3 in the magazine Nature.
The updated system is able to work with large structures such as DNA and RNA, carrying genetic code, as well as with much smaller particles – ions of metals.
Alphafold 3 development was carried out by DeepMind and ISOMORPHIC Labs – “nursing” company Alphabet, which is working on and tools for biotechnology under the guidance of the same Hassabis. In January, ISOMORPHIC Labs announced partnership with pharmaceutical giants Eli Lilly and Novartis for the joint development of drugs using neural networks.
Unlike previous open versions, the source code of Alphafold 3 will not be published in the overall access. However, DeepMind will provide free access to the cloud system for external researchers. As John Jamper, who heads the developer team, noted, this will also make it possible to find out how proteins interact with DNA inside the body. “How do proteins react to DNA damage, how do they detect and restore these damage? Now we can answer such questions,” he said.
Previously, to determine the structure of proteins, painstaking work using electronic microscopes and the method of radiographic analysis was required. Several years ago, groups of researchers began to find out whether the deep training underlying many recent breakthroughs in AI was able to precisely predict the form of proteins exclusively in their components of their amino acids, studying at structures that were verified experimentally.
In 2018, Google Deepmind said she was working on Alphafold. In 2020, AlphaFold 2 demonstrated the