Google DeepMind releases AI model of living cells

Google DeepMind unveiled a model of artificial intelligence that explains the interactions of cells and life’s building block. This will help unlock disease secrets and develop treatments for cancer and other diseases.

AlphaFold 3 is the third generation technology developed in 2018. It gives the most sophisticated predictions yet on how small biological structures will look and mix, according to an article published in Nature Wednesday.

The model was developed in collaboration with Isomorphic Labs a DeepMind spin-off that specializes in drug discovery. It is the latest milestone on the journey to use AI’s predictive power to better understand life’s tiny mechanisms.

“Biology, like all dynamic systems, is dynamic. You have to understand the properties of biology that emerge from the interactions between molecules within the cell,” said Sir Demis Hazzabis. He is DeepMind’s co-founder and chief executive. AlphaFold 3 is our first major step in that direction.

The new technology expands the scope of its analysis beyond proteins, allowing a more comprehensive view of biochemical networks which make organisms work. The model includes the genetic code, DNA and RNA, as well as ligands – molecules that can bind with others and be important markers for disease.

Max Jaderberg is the chief AI officer at Isomorphic Labs. He said that AlphaFold 3 offers researchers new opportunities to identify new drug molecules quickly. Isomorphic Labs is partnered with Eli Lilly and Novartis.

Jaderberg explained that AlphaFold 3 allows scientists to test hypotheses on an atomic scale and produce accurate structure predictions in seconds. This is in comparison to months or years that it would take to perform this experimentally.

The paper claims that AlphaFold 3 has “significantly improved’ predictive accuracy compared to many other specialised tools, including those built on its predecessors. The research shows that the development of the best AI Deep Learning frameworks can reduce the amount data required to achieve “biologically-relevant performance”.

John Jumper said that the AlphaFold technique could improve plant biology, and therefore food security. We’re starting to see early testers and biologists use this technique to understand how the cells work — and begin to think about what might go wrong in disease states.

AlphaFold 3’s suggested molecules will need to go through the usual process of clinical trials and be tested experimentally. DeepMind has announced that it will make the AlphaFold 3 functionality freely available to academic users.

Boston Consulting Group’s study, published this week, suggests that AI-discovered drugs have a greater success rate than drugs discovered using other methods in early stage trials. Researchers cautioned that this was a preliminary analysis of AI’s efficacy in drug discovery. They said AI could double productivity of pharmaceutical research and development.

The server is expected to revolutionize the way scientists conduct experiments. This was stated by Julien Bergeron of King’s College London’s structural biology department, who wasn’t involved in its development but was a beta tester.

He said, “We can test hypotheses even before we go to the laboratory.” This will be a game changer.

AlphaFold 3 has limitations, including the inability to deal with mirror-image molecules and “hallucinations” that “spurious structure order” is present in areas which are actually disordered. The model assigns confidence measures to its predictions to reflect the probability of error.