An illustration of antibodies (pink) binding to a flu virus (brown).Credit: Juan Gaertner/Science Photo Library
Biologists achieved a landmark in protein design last year: using artificial intelligence (AI) to draw up entirely new antibody molecules1. Yet the proof-of-principle designs lacked the potency and other key features of commercial antibody drugs that rack up tens of billions in annual sales.
After a year of progress, scientists say they are on the cusp of turning AI-designed antibodies into potential therapies. In recent weeks, numerous teams have reported successfully using proprietary commercial AI tools and open-source models to make antibodies that have properties of antibody drugs.
‘A landmark moment’: scientists use AI to design antibodies from scratch
“These latest efforts are remarkably powerful advances that enable a democratization of antibody engineering,” says Chang Liu, a synthetic biologist at University of California, Irvine.
The latest wave of success with de novo antibody design “will have a big impact on how quickly and how many de novo therapeutics we will see in clinical trials”, adds Timothy Jenkins, a protein engineer at the Technical University of Denmark in Kongens Lyngby.
Precision nanobodies
Therapeutic antibodies are usually made by screening vast numbers of diverse antibodies to find ones that can recognize a certain target. But sometimes, these screens turn up only antibodies that bind weakly to the target or recognize the wrong region on it.
“There isn’t much precision,” says Surge Biswas, chief executive of antibody-design company Nabla Bio in Cambridge, Massachusetts. Instead, scientists hope to specify an antibody’s desired target — the active site of an enzyme implicated in disease, for instance — and have an AI model suggest designs. “The promise of AI-guided design is that you can be atomically precise,” Biswas adds.
AlphaFold is five years old — these charts show how it revolutionized science
... continue reading