Google and its DeepMind AI unit six years ago unveiled AlphaFold, an AI model designed to predict protein models, and two years later rolled out AlphaFold 2, which offered significant improvements in the model’s predictions.

This week, DeepMind and its sister business, Isomorphic Labs, introduced AlphaFold 3, the latest iteration that goes well beyond the capabilities of the first two to not only predict the structures of such biological molecules as proteins, DNA, RNA and ligands but also how these biomolecules will interact with each other, a landmark step in the discovery of novel drugs that also will have uses in such areas as genomics, biorenewable materials, and the development of more resilient crops to ensure global food security.

In a blog post and a paper published in Nature, the AI researchers said the new AI model is at least 50% more accurate than existing prediction models when used to understand the interaction of proteins with other molecule types and twice as accurate in some key categories of interaction.

Google also is making AlphaFold 3 and its database of 200 million protein structures free to scientists for non-commercial use through its new AlphaFold Server, promising to accelerate the development of novel drugs, improve their chances of success and reduce the costs. Meanwhile, Isomorphic Labs – an AI unit Google created in 2021 to focus on drug discovery – is working with pharmaceutical companies to help them use AlphaFold 3 for their internal projects.

“AlphaFold 3 brings the biological world into high definition,” the DeepMind and Isomorphic Labs researchers wrote. “It allows scientists to see cellular systems in all their complexity, across structures, interactions and modifications. This new window on the molecules of life reveals how they’re all connected and helps understand how those connections affect biological functions – such as the actions of drugs, the production of hormones and the health-preserving process of DNA repair.”

Pharmaceutical Industry’s Embrace of AI

The massive pharmaceutical industry – which market research firm Acumen Research said was an $81.5 billion market in 2022 and will grow to $181.4 billion by 2032 – has been a heavy adopter of AI technologies, according to the World Health Organization.

“AI is already used in most steps of pharmaceutical development, and, in the future, it is likely that nearly all pharmaceutical products that come to market will have been ‘touched’ by AI at some point in their development, approval or marketing,” WHO wrote in March. “Although these uses of AI may have a commercial benefit, it is imperative that use of AI also has public health benefit and appropriate governance.”

Thought AI isn’t anything new to pharmaceutical companies, the introduction of generative AI into the conversation in November 2022 with OpenAI’s ChatGPT gave the industry a shot in the arm, Ben Mabey, CTO of drug discovery company Recursion Pharmaceuticals, told BioSpace – a community hub for life sciences news and jobs – in an article this month.

“[That] triggered the imaginations of execs across the board.” Mabey said. “It’s just an exciting time to be in the space now because people are finally waking up to the potential here.”

AlphaFold’s Role in Drug Discovery

Google is expecting AlphaFold 3 and subsequent versions to play a significant role the pharmaceutical industry’s core drug discovery role.

“As a single model that computes entire molecular complexes in a holistic way, it’s uniquely able to unify scientific insights,” the AI units wrote. “AlphaFold 3 creates capabilities for drug design with predictions for molecules commonly used in drugs, such as ligands and antibodies, that bind to proteins to change how they interact in human health and disease.”

They added that the “the ability to predict antibody-protein binding is critical to understanding aspects of the human immune response and the design of new antibodies – a growing class of therapeutics.”

AlphaFold 3 also can model chemical modification to the molecules, which control the healthy functioning of cells and that, when disrupted, can lead to disease.

The model, which covers all of life’s molecules, uses a list of those molecules and then generates a 3D structure that shows how they fit together. It leverages an improved version of DeepMind’s Evoformer module, a deep learning architecture that was foundational to AlphaFold 2. After processing the molecule list, AlphaFold 3 uses a diffusion network similar to those used by AI image generators to pull together its predictions.

“The diffusion process starts with a cloud of atoms, and over many steps converges on its final, most accurate molecular structure,” the researchers wrote.

AlphaFold Server Speeds Up Research

AlphaFold Server lets scientists use the AI model to speed up their work and reduces the amount of time needed to run their predictions. It also levels the playing field for these scientists, many of whom may not have access to the compute capabilities needed to run their tests or the expertise to develop and run machine learning environments.

Running experiments to predict protein structures can cost hundreds of thousands of dollars and take years to complete, according to DeepMind, which Google bought in 2014 and last year merged with Google Brain AI unit to create Google DeepMind. Researchers have used AlphaFold 2 to predict hundreds of millions of structures in work that would have taken hundreds of millions of years with existing tools, the AI unit wrote.

In a statement, Céline Bouchoux, principal laboratory research scientist at The Francis Crick Institute, a London-based biomedical research center, said that “with AlphaFold Server, it’s not only about predicting structures anymore. It’s about generously giving access: Allowing researchers to ask daring questions and accelerate discoveries.”