Scientists use AI to create antibodies entirely from scratch

Researchers from Nobel Laureate David Baker’s Lab and the University of Washington’s Institute for Protein Design (IPD) have achieved a major milestone by using artificial intelligence to design antibodies from scratch – a development that could transform the $200 billion antibody drug industry.

This new research, published in Nature, is another key breakthrough for Baker’s team, whose previous work on AI-powered protein design won him last year’s Nobel Prize in Chemistry.

A ‘pipe dream’ realised

“It was really a grand challenge – a pipe dream,” said Andrew Borst, Head of Electron Microscopy R&D at IPD. Now that the team has succeeded in engineering antibodies that successfully bind to their targets, Borst said the research “can go on and it can grow to heights that you can’t imagine right now.”

Until now, antibody discovery rested upon on immunising animals and waiting to see if useful molecules were produced – a slow, costly and unpredictable process. Despite these challenges, this research has been fundamental to the development of new treatments for cancer and autoimmune diseases.

Designing the building blocks of life

Baker’s Nobel-winning work focused on deciphering the molecular design of proteins and creating AI tools that could model and refine them in silico. The new technology learns from existing proteins and how they function, then proposes entirely new molecular designs to tackle specific biological problems.

In this latest research, researchers turned their attention to the six flexible loops of protein on an antibody’s arms – the structures that grip onto targets such as viruses or toxins.

The new technology learns from existing proteins and how they function, then proposes entirely new molecular designs to tackle specific biological problems.

“We are starting totally from scratch – from the loop perspective – so we’re designing all six,” explained Robert Ragotte, a postdoctoral researcher at IPD. “But the rest of the antibody, what’s called the framework, that is actually staying the same.”

By retaining much of the antibody’s human framework, the team hopes to reduce the likelihood of triggering immune rejection in patients while still creating highly specific and potent molecules.

Promising laboratory results

To test their AI-generated designs, the researchers challenged them against a range of real-world targets, including hemagglutinin, a protein on flu viruses that enables infection and a toxin produced by the C. difficile bacteria.

To test their AI-generated designs, the researchers challenged them against a range of real-world targets.

The lab results closely matched the AI’s predictions. “They were binding in the right way with the right shape against the right target at the spot of interest that would potentially be useful for some sort of therapeutic effect,” Borst said. “This was a really incredible result to see.”

Thanks to the close collaboration between computational scientists and experimental biologists, the team was able to quickly refine their digital designs based on the findings in the lab.

Open tools and future prospects

The software used for the antibody design has been made freely available on GitHub, allowing other scientists to build upon the work. Meanwhile, Xaira Therapeutics, a biotech start-up founded by IPD alumni, has licensed parts of the technology for further commercial development.

Despite the success, the researchers emphasise that the designed antibodies are only the beginning. Further optimisation will be needed to enhance solubility, target affinity and reduce the risk of unwanted immune responses before any potential therapies can reach patients.