One of this year’s Nobel Prize winners in physics, Geoffrey Hinton, who pioneered work on the neural networks that undergird artificial intelligence, has warned that machines might someday get smarter than humans. Perhaps. But this year’s Nobel Prize in chemistry honored a real-world example of how AI is helping humans today with astounding discoveries in protein structure that have far-reaching applications. This is a development worth savoring.
Proteins are biology’s lead actors. As the Nobel committee pointed out, proteins “control and drive all the chemical reactions that together are the basis of life. Proteins also function as hormones, signal substances, antibodies and the building blocks of different tissues.” In the human body, they are necessary for the structure, function and regulation of tissues and organs. All proteins begin with a chain of up to 20 kinds of amino acids, strung together in a sequence encoded in DNA. Each chain folds into a unique structure, and those shapes determine how proteins interact with other molecules.
Looking like a tangled ball of twine, proteins have a complex and precise design of moving parts that are linked to chemical events and bind to other molecules. Antibodies are proteins produced by the immune system that bind to foreign molecules, including those on the surface of an invading virus, such as the spikes on the coronavirus that causes covid-19.
At the end of the 1950s, University of Cambridge researchers John Kendrew and Max Perutz successfully used a method called X-ray crystallography to produce the first 3D models of proteins. In recognition, they were awarded the 1962 Nobel Prize in chemistry. In the ensuing half-century, the quest to document protein structures remained laborious and slow. A single protein structure might take a doctoral student four or five years to figure out. Before AI, the field’s central repository contained some 185,000 experimentally solved protein structures.
This year’s Nobel Prize in chemistry went to three scientists who revolutionized the field. David Baker of the University of Washington built entirely new kinds of proteins. Demis Hassabis and John Jumper of DeepMind, a Britain-based firm that is part of Alphabet, Google’s parent company, developed an AI and machine learning model that can predict the structure of proteins, decoding the amino acids that make up each. The model, AlphaFold, can do in minutes what once took years.
AlphaFold takes advantage of neural networks that can locate patterns in enormous amounts of data. The system was trained on the vast information in the databases of all known protein structures and amino acid sequences. AlphaFold has predicted more than 200 million protein structures, or nearly all catalogued proteins known to science, including those in humans, plants, bacteria, animals and other organisms. The AlphaFold Protein Structure Database makes this data freely available.
To design new drugs and vaccines, scientists need to know how a protein looks or behaves. The AlphaFold result is a prediction — a visual representation of a protein’s expected structure — but such predictions can accelerate biomedical research.
The AlphaFold blog recounts the story of scientists searching for a better vaccine against malaria, a disease that afflicts 250 million people a year and causes more than 600,000 deaths. Because malaria is caused by a shape-shifting parasite, vaccine researchers had long struggled to characterize the structure of one surface protein they needed to target to interrupt the infection. Then AlphaFold’s prediction of the right structure snapped it into focus. Matthew Higgins at the University of Oxford said the breakthrough helped his team decide which bits of the protein to put in the vaccine, which trains the body’s immune system to detect it and act. This helped advance his research from “a fundamental science stage to the preclinical and clinical development stage.”
Anyone who has used ChatGPT knows that AI is not always correct — and the malaria scientists found that some of the 3D visualizations of proteins were inexact. But AI will only get better over time. Already, the AlphaFold effort is expanding to create accurate visualizations of how proteins interact with other biomedical structures, such as nucleic acids.
In the years ahead, AI dangers must be confronted and safeguards considered. Without a doubt, there are risks when powerful technology falls into the hands of malign actors.
But, for now, AlphaFold shows that AI can supercharge existing knowledge to benefit mankind. The Nobel committee noted that, thanks to these advances, “researchers can now better understand antibiotic resistance and create images of enzymes that can decompose plastic.” And there will be more to come.
The Washington Post
