AI transforms the way we do science: cracking the secret code of proteins, for example

The 2024 Nobel Prize in Chemistry is about proteins, life’s ingenious chemical tools. 

It was awarded on Wednesday October 9, 2024, to three scientists for discoveries that show the potential of advanced technologies, including artificial intelligence, to predict the shape of proteins, and invent new ones.

The diversity of life testifies to the astonishing capacity of proteins as chemical tools. They drive and control all the chemical reactions that are the basis of life. They serve as hormones, signaling substances, antibodies and building blocks of various tissues and enzymes. They influence the behavior and growth of viruses and bacteria. 

David Baker of the University of Washington used computer software to invent a new protein. 

Demis Hassabis and John Jumper of Google DeepMind developed an AI model to solve a 50-year-old problem: predicting protein’s complex structures in 3D from their amino acid sequences.

“Both of these discoveries open up vast possibilities,” said Heiner Linke, Chair of the Nobel Committee for Chemistry.

Demis Hassabis and John Jumper

Initially, proteins appear as chains of chemical compounds, before twisting and folding into three-dimensional shapes that define what they can and cannot do. For many years, it has been laborious to determine the precise shape of individual proteins, and for over 50 years scientists have struggled to solve what was called “the protein folding problem”.

In a 2022 interview with Scientific American, Hassabis stated, “AI appears to understand the diverse forces that attract and repel amino-acid components to and from one another. Such forces can move and twist a protein into specific configurations that are predictable by AI. These predictions allow researchers to explore how proteins could be used to develop new pharmaceutical products, for example”.

AlphaFold is built using a mathematical system known as a neural network. It can predict the shape of a protein in the human body, determining how other molecules will bind or physically attach to it. Understanding how a drug binds to particular proteins in the body, and altering their behavior, is one way of developing new drugs. Then in 2020, with the unveiling of an update to the technology, AlphaFold2, Google researchers proved that the technology could predict shapes with a level of accuracy rivaling that of physical experiments.

David Baker

David Baker’s work preceded the emergence of the latest A.I. models and focused on protein creation.

In 2003, Dr. Baker and his colleagues created the first entirely new protein: a molecule called Top7, that was useless but emblematic.

“Until then, really the only proteins that were known were those derived from millions or billions of years of evolution,” he said in an interview with The New York Times.

Starting with the desired protein shape, they used a computer model called Rosetta, which searches existing protein databases for the amino acid sequence that might create the desired structure.

Dr. Baker recalls the “incredible moment” when the protein he had created from bacteria, based on the proposed amino acid sequence, presented virtually the same structure as his model. 

In recent years, his laboratory has used neural networks not only to predict the shapes of proteins, but also to generate blueprints for new proteins. He realized that if he could create a novel protein structure; he should also be able to create more sophisticated proteins “capable of real prowess”, like breaking up the amyloid fibrils that are thought to be involved in Alzheimer's disease!

Why is this discovery so important?

Life could not exist without proteins. The fact that we can now predict protein structures and design our own proteins represents a huge step forward for mankind.

“To understand how proteins work, you need to know what they look like, and that's what this year's laureates have done,” said Johan Aqvist of the Nobel Committee for Chemistry.

The key to understanding protein function is discerning how these long polymers wrap themselves up in space. These wobbly 3D shapes determine the arrangement of their chemically active side chains and the places where other molecules might interact with the protein. Now, thanks to the researchers awarded the 2024 Nobel Prize, scientists are able to predict those arrangements and design new proteins for specific functions. 

The task of understanding how proteins fold in 3D used to take months, even decades. But AI models like AlphaFold make it possible to do so in hours, or even minutes.