This new approach is called Neural Iterative Selection Expansion (NISE), and rather than experimentally screening and rejecting thousands of protein candidates, the AI proposes a protein structure, predicts how tightly it will bind to a chosen drug, and continuously refine this design until it identifies candidates with enough potential to warrant lab testing.

The researchers first applied NISE to exatecan, a highly potent chemotherapy drug used in antibody-drug conjugates. The AI-designed protein was able to wrap around the drug, protecting a chemically fragile region that would’ve otherwise degraded. In future, similar proteins could help stabilise cancer medicines, deliver them more precisely to tumours, monitor where drugs travel in the body or even capture toxic drugs that are accidentally released into healthy tissue.

In fact, just to demonstrate the software’s versatility, the researchers also designed a protein capable of binding apixaban, a widely used blood thinner. Though still at an early stage, the team is confident this approach could eventually lead to antidotes that rapidly neutralise drugs when serious side effects occur.

While this technology has not yet progressed beyond laboratory studies, it demonstrates how AI can dramatically reduce the time required to engineer highly specified proteins, opening new possibilities for precision medicine and next generation cancer treatment.

“Although the technology is still at an early stage, it could ultimately support safer drug delivery, better measurement of drug exposure and new ways to limit toxicity.” Dana-Farber Cancer Institute