Google DeepMind has released a new version of AlphaFold, a landmark tool for predicting protein structures, that puts the artificial intelligence software on a path to make breakthroughs in biology research and bolster a business that Google’s AI chief says could be worth north of $100 billion.
The AI system has the potential to revolutionize medicine and create “enormous commercial value,” for DeepMind spinout Isomorphic Labs, Demis Hassabis, the chief executive officer of both subsidiaries, said in an interview with Bloomberg Television. “I hope to do both with Isomorphic: build a multi-hundred billion dollar business — I think it has that potential — as well as be incredibly beneficial for society and humanity.’’
Isomorphic Labs, a unit of Alphabet Inc. created three years ago, was built to commercialize DeepMind’s AI for drug discovery. DeepMind first released AlphaFold in 2018, with advances in decoding the shape of proteins, a scientific problem often compared to mapping the human genome. Now in its third iteration, AlphaFold can model a range of molecular structures, including DNA and RNA, and predict how they interact with one another.
“To really understand biology, you need to think about the interactions between different biological molecules,” Hassabis said. “And that's what AlphaFold 3 is a big step towards.”
Hassabis described the update, which was also published in the scientific journal Nature, as “essential for drug discovery” because of its critical insight into the kinds of chemical compounds used in designing and testing new medicine, including vaccines. “I would be expecting, maybe in the next couple of years, the first AI-designed drugs in the clinic,” Hassabis added.
AI in medicine is a hot field. It’s driven by a belief that software algorithms can find and help develop new drugs at drastically lower costs and much faster than the decades it can take now. The craze has drawn pharmaceutical firms, venture capitalists and tech giants like Nvidia Corp. into a market worth an estimated $50 billion. Investors have poured more than $18 billion into “AI-first” biotech companies in the past decade.
But even with more cash and computing advances, companies working on the tech have yet to show major clinical success. Some biotech firms specializing in AI drug discovery, such as BenevolentAI and Exscientia Plc, have struggled in public markets.
In January, Isomorphic Labs announced its first two commercial partners, Eli Lilly & Co. and Novartis — alliances the tech company said “could be worth nearly $3 billion” if they’re successful on a range of performance milestones.
Alphabet has tried turning its health-care innovations into new business opportunities before, with units such as Verily, a “precision health” platform for managing medical data. But those have shown limited commercial success so far.
Hassabis has previously said that Google will spend more than $100 billion to develop its AI arsenal. “The benefits of the generative AI models, things like drug discovery, are going to far outweigh these costs in the long run,” he said in the interview this week. DeepMind is responsible for some of Google’s biggest advances in AI research. Since merging with Google’s other AI units last year, DeepMind now leads the company’s efforts on Gemini, its foundational AI model competing with OpenAI.
To create AlphaFold 3, Hassabis said his researchers rebuilt the program “from the ground up,” using methods that account for shortcomings in prior versions, like a deficit of training data. The new version relies on an AI technique used in generative AI, called a diffusion model, that allows computers to do things like turn text into hyper-realistic images. In the Nature paper, DeepMind researchers write that their model predicts certain molecular interactions — proteins with nucleic acid and antibodies with antigens — at a higher accuracy than existing computer models. “It’s a lot more generalizable,” Hassabis said of the new model. “And it’s a lot more powerful.”
Along with the update, DeepMind is launching AlphaFold Server, a tool that gives outsiders access to a bulk of the program’s functions. Hassabis said millions of researchers have used AlphaFold so far, although they “have to be a little bit technical” to deploy the program effectively. “The Server is even simpler,” he said. “If you’re a biologist, you don't need to understand the model or the technology at all.”
DeepMind will offer the tool to scientists for “non-commercial use.” Although Hassabis added that Isomorphic Lab’s pharmaceutical partners will have access to the Server and other more advanced tools.
A message from Advisor Perspectives and VettaFi: To learn more about this and other topics, check out some of our webcasts.
Bloomberg News provided this article. For more articles like this please visit
bloomberg.com.
Read more articles by Mark Bergen, Tom Mackenzie
