Three scientists were named winners of the 2024 Nobel Prize in Chemistry for their innovations in the fields of computational protein design and structure prediction. One half of the prize was awarded to David Baker, PhD, from the University of Washington for computational protein design and the other half was awarded jointly to Sir Demis Hassabis, PhD, and John M. Jumper, PhD, from Google DeepMind for their work on computational protein structure prediction using AlphaFold.
The discoveries honored by this year’s award can trace their roots back about 65 years when the earliest three-dimensional protein structures were determined using X-ray crystallography. Determining protein structures experimentally is extremely labor intensive due to the sheer number of theoretically possible three-dimensional conformations that proteins can take. In fact, early predictions based on calculating probabilities were not very accurate.
In the late 1980s, scientists took some of the first steps towards de novo protein design but constructing complex proteins, such as those with mixed topologies, remained a grand challenge for the community.
The real breakthrough came in 2003 when Baker and his colleagues solved the protein design problem using Rosetta, a computer program for protein structure prediction and design that they had developed in 1999. The team predicted the structure of Top7, a 93-residue α/β-protein, and validated their design. Baker’s work made various kinds of studies possible including research into things like virus inhibitors, protein-based fentanyl sensors, nanomaterials, and much more.
Meanwhile, scientists continued trying to improve the accuracy of protein structure prediction tools. Efforts to move the field forward include the launch of the Critical Assessment of Protein Structure Prediction (CASP) challenges in 1994. Founded by John Moult and Krzysztof Fidelis, these challenges encouraged scientists to use their best methods to make predictions about unknown protein structures. Their results were compared to the structures of those proteins determined by X-ray crystallography. The methods were ranked based on how closely the predicted structures matched the true structure.
While these challenges were certainly instrumental in moving the field forward, the accuracy of the predictions remained low for a number of years. Things changed dramatically in 2018, with the launch of AlphaFold developed by Hassabis and Jumper, which uses neural networks and deep learning algorithms to learn from databases of protein structures and then make predictions about new structures. By 2020, with the release of AlphaFold2, it was clear that Hassabis, Jumper and their team had successfully solved the protein structure prediction problem for monomeric proteins.
Today, AlphaFold and later versions of the program have been used in studies done by more than two million users worldwide. For example, last year, scientists from Stanford University, the University of Toronto, and Insilico Medicine used AlphaFold in their efforts to identify a treatment for a novel target in cases of hepatocellular carcinoma, the most common form of primary liver cancer. In May of this year, scientists from the University of North Carolina, School of Medicine, Stanford, Harvard, and the University of California, San Francisco, published a study that assessed AlphaFold2’s effectiveness at both producing 3D models of proteins and predicting ligand activity. Their findings suggested that AlphaFold2 structures could indeed be useful for drug discovery.
Meanwhile, AlphaFold2 has also been used to predict structures for about 200 million proteins all of which are stored in a database that can be mined for things like plastic-degrading enzymes and to better understand antibiotic resistance. The DeepMind team also published a paper in Nature this year that shared details of the expanded predictive capabilities available in AlphaFold3.
In addition to the Nobel, AlphaFold’s developers also shared the 2023 Breakthrough Prize in life sciences.
Of the three winners, only Baker was able to join the press conference announcing the award this morning. He described himself as both “very excited and very honored” to receive the award. In response to questions from the press, he was quick to highlight the work of other scientists in the protein structure space that were instrumental to his work including the work of Dahiyat and Mayo who published the first successful design of a small protein using computation in 1997 stating, “I stood on the shoulders of giants.” He also acknowledged the contributions of his various colleagues and collaborators.
Baker also highlighted the breakthroughs in AI and the development of AlphaFold by Hassabis and Jumper. Their work “really highlighted to us the power that AI could have and that led us to apply these AI methods to protein design. These AI methods are much more powerful than traditional scientific model methods. I am really excited about all the ways in which protein design can now make the world a better place.”
Baker will be a keynote speaker at GEN’s inaugural summit on the State of AI in Drug Discovery which will be held on October 30.