Photo: PublicDomainPictures / Pixabay
For half a century, working out how a protein folds took years of painstaking lab work — and there are billions of them. Then an AI learned to predict the answer in seconds. It has now mapped nearly every known protein, and it's starting to design medicines.
Every living thing runs on proteins. They digest your food, carry oxygen in your blood, fight off infections, and build the structure of your cells. And a protein’s job is decided almost entirely by one thing: its shape.
The catch is that a protein is built as a long chain that then folds itself, in a fraction of a second, into an intricate three-dimensional knot. Figuring out that final shape — from the chain alone — was one of the hardest open problems in all of biology. For 50 years it resisted the world’s best scientists.
Then an AI solved it. And in 2024, that achievement won the Nobel Prize in Chemistry.
Why folding was so hard
Here’s the problem in a sentence: a protein could theoretically fold into an astronomical number of possible shapes, but in reality it snaps into just one. Predicting which one, from the chemical recipe, is fiendishly complex.
For decades, the only reliable way to find a protein’s structure was to determine it experimentally — painstaking lab techniques that could take a PhD student years to nail down a single one. Meanwhile, biology had sequenced the recipes for hundreds of millions of proteins whose shapes remained a mystery. The gap between “recipes we know” and “shapes we understand” was a canyon.
Enter AlphaFold
The system that closed the gap is called AlphaFold, built by the AI lab Google DeepMind. Rather than simulating the physics of folding, it learned. Trained on the proteins whose structures scientists had painstakingly solved, it taught itself the deep patterns connecting a chain to its final shape.
The result stunned the field. At a long-running contest where teams compete to predict protein structures, AlphaFold didn’t just win — it produced predictions so accurate they rivalled the slow, expensive lab experiments. A problem measured in years collapsed to one measured in seconds.
Nearly every protein we know, mapped
DeepMind then did something that multiplied the impact enormously: it ran AlphaFold on essentially every protein science had on file — over 200 million of them — and released the entire database, free, to the world.
Overnight, researchers who might have spent years chasing a single structure could look it up. That database has since been used by millions of scientists in nearly every country on Earth, feeding work on everything from antibiotic resistance to neglected tropical diseases to the enzymes that might one day break down plastic waste.
In 2024, the Nobel Prize in Chemistry recognised this work — shared between the DeepMind researchers behind AlphaFold’s prediction and a scientist honoured separately for the equally remarkable feat of designing entirely new proteins from scratch.
From reading proteins to designing drugs
Knowing a protein’s shape isn’t just academic. Most modern medicines work by latching onto a specific protein — a lock that the drug is the key for. If you can see the lock in detail, you can design a better key.
Newer versions of the system go further, predicting not just a protein’s shape but how it will interact with other molecules — exactly the question a drug designer needs answered. It’s turning the slow, luck-heavy early stages of drug discovery into something faster and more deliberate.
The bigger lesson
It’s tempting to lump AlphaFold in with chatbots and image generators, but it’s a different — and in some ways more important — story about AI. This isn’t a machine producing plausible words. It’s a machine that solved a genuine, decades-old scientific problem, produced answers experimentalists could verify, and handed the results to the whole world for free.
That’s the version of AI worth getting excited about: not the one that writes your emails, but the one quietly rewriting what’s possible in a laboratory. A problem that stumped biology for half a century is, essentially, solved. And the same approach is now being pointed at cancer, at new antibiotics, at diseases that have resisted us for generations.
The machines learned to read the language of life. We’re only beginning to find out what they’ll help us write with it.
Sources & further reading
Researched and written with the help of AI tools and edited for accuracy. Provided for general information and discussion only — not professional advice. See our editorial standards and disclaimer. Spotted an error? Tell us.
Enjoyed this? Get the next one.
One good read at a time, straight to your inbox. No spam, unsubscribe anytime.
