Artificial Intelligence and Improved Drugs: Another Big Step Forward

Designing drugs that target proteins requires knowing the three-dimensional (3D) shape of the target protein and creating a drug whose shape binds to that target protein. As of June 2021, nearly a century of research had determined the shape of only about one third of human proteins. Then, in July 2021, two groups reported using artificial intelligence (AI) to predict the shape of nearly every human protein from their nucleic acid sequences (NEJM JW Gen Med Sep 1 2021 and Nature 2021; 596:590 and 583). In an editorial, Science called it the scientific breakthrough of the year. Yet skeptics hypothesized that new drugs based on this knowledge would be far in the future.

The same AI software that predicted the shape of proteins in the 2021 experiments now has been used to design small strings of 50 to 65 amino acids (“mini-antibodies”) with 3D shapes that were predicted to bind to 13 important human target proteins: growth factors that are important in stimulating growth of cancers, and components of several infectious agents, including influenza and SARS-CoV-2. The mini-antibodies were shown to bind to their targets and to work as expected — for example, a mini-antibody designed to bind to the SARS-CoV-2 spike protein protected mice from infection.