Google DeepMind’s latest medical breakthrough borrows techniques from artificial intelligence image generator

Much of the recent AI hype has focused on mesmerizing digital content generated by simple prompts, but also on concerns about its ability to massively reduce labor and make malicious propaganda more convincing. (Interesting!) However, the most promising (and perhaps less ominous) jobs for artificial intelligence lie in medicine. A new update to Google’s AlphaFold software could lead to new breakthroughs in disease research and treatment.

The AlphaFold software from Google DeepMind and Isomorphic Labs (also owned by Alphabet) has shown that it can predict how proteins fold with astonishing accuracy. It documents a staggering 200 million known proteins, and Google says millions of researchers have used previous versions to make discoveries in areas such as malaria vaccines, cancer treatments and enzyme design.

Understanding a protein’s shape and structure determines how it interacts with the body, allowing scientists to create new drugs or improve existing ones. But the new version of AlphaFold 3 can simulate other key molecules, including DNA. It can also map interactions between drugs and diseases, which could open exciting new doors for researchers. Google says it’s 50% more accurate than existing models.

“AlphaFold 3 allows us to scale from proteins to a broad range of biomolecules,” the Google DeepMind research team wrote in a blog post. “This leap could unlock more transformative science, from developing biorenewable materials to more resilient crops, to accelerate drug design and genomics research.”

“How do proteins respond to DNA damage? How do they find and repair it?” said John Jumper, Google DeepMind project leader wired. “We can begin to answer these questions.”

Before the advent of artificial intelligence, scientists could only study protein structures through complex methods such as electron microscopy and X-ray crystallography. Machine learning simplifies much of this process by using patterns identified from training (often imperceptible to humans and our standard instruments) to predict protein shape from amino acids.

Google says part of AlphaFold 3’s progress comes from applying diffusion models to its molecular predictions. The diffusion model is a core part of AI image generators such as Midjourney, Google’s Gemini, and OpenAI’s DALL-E 3. wired explain. In other words, it takes a formation that looks vague or fuzzy and clears it by making educated guesses based on patterns in the training material.

“This is a huge step forward for us,” said Demis Hassabis, CEO of Google DeepMind. wired. “That’s exactly what drug discovery requires: You need to understand how a small molecule binds to a drug, how strongly it binds, and what else it might bind to.”

AlphaFold 3 uses a color-coded scale to mark the confidence of its predictions, allowing researchers to exercise appropriate caution against results that are unlikely to be accurate. Blue indicates high confidence; red indicates less sure.

Google is making AlphaFold 3 free for researchers to use for non-commercial research. However, unlike past versions, the company has not open sourced the project.David Baker, a professor at the University of Washington and a well-known researcher who has created similar software, is interested in wired Google chose this route. However, he was also surprised by the capabilities of the software. “AlphaFold 3’s structure prediction performance is very impressive,” he said.

As for next steps, Google said, “Isomorphic Labs is already working with pharmaceutical companies to apply it to real-world drug design challenges and ultimately develop life-changing new treatments for patients.”