How Is AI Designing New Medicine?

A team of scientists at the University of Washington has used artificial intelligence to build new tools for treating diseases that involve shapeless proteins. These proteins, called intrinsically disordered proteins, don’t have a fixed structure. That has made them nearly impossible to target using standard drugs.

Led by Nobel Prize winner David Baker, the group built tiny custom-made proteins, known as binders, using machine learning. These binders were made to stick to the slippery targets and block their effects. Some of these targets are linked to cancer, diabetes, pain, and brain conditions.

The research team tested their method on 43 targets and succeeded with 39 of them. That’s according to a study published in Science on 17 July 2025. One binder stopped pain signals in lab-grown human cells. Another broke down clumps of proteins tied to type 2 diabetes.

Baker described the method as a new tool for handling a large group of human proteins that researchers have struggled to reach before.

 

What Makes These Proteins So Difficult?

 

Normal proteins fold into clear shapes and these shapes give them fixed places where drugs can latch on. Many treatments work this way, like keys fitting into locks. But disordered proteins behave more loosely and are constantly moving.

This shapelessness means they don’t have a clear point of contact. Standard drugs and even antibody methods have failed against them. And that’s a problem, since these proteins play a part in many diseases.

Kejia Wu is one of the lead authors of the study and she said it was hard to know where to start. But their loose structure gave her and her colleagues more room to try ideas. Most of them didn’t work, but the ones that did, worked well. Wu said this gave her freedom to think in new ways.

In fact, she said that the same thing that made these proteins hard to deal with also made them easier to catch from other angles. If there are more ways to grab something, you only need one of them to succeed.

 

 

How Does The AI Design New Binders?

 

The process started with a library of protein parts. These were stitched together in many combinations. Then, using a technique called diffusion modelling, the AI tweaked these pieces to better fit the targets.

Each new design was tested in the lab using a method called biolayer interferometry. This helped the scientists see if their binder latched on to the right target—and only that target.

In the test on dynorphin A, a protein tied to pain response, the AI-built binder locked on better than the protein’s usual partners. This allowed the scientists to block pain signals in a way that hadn’t been done before.

The research team didn’t stop at one type of target. Their method worked on a wide mix of proteins that have long caused problems for scientists. Each design was made from scratch, based on what the AI predicted would work best.

 

Could This Method Change How Medicine Ts Built?

 

The binders could help in more than one way. Many health problems involve proteins that don’t behave in a steady or predictable way. These new binders may let researchers quiet those proteins, trace where they go in the body, or learn what they do.

There’s also interest in using them for studying parts of the cell that were hard to look at before. For example, cells contain floating blobs called condensates that carry out tasks like gene control and immune response. These blobs rely on disordered proteins. The new binders could help scientists work with those blobs directly.

The tools are already available online. Other researchers can try the designs in their own labs. The next step will be to test how long these binders last and how safe they are to use in real life.

Wu said the work took patience. “Most of your ideas will fail,” she said. “But then one of them works, and that’s the part I like.”