Tara Chalasani | January 8th, 2025
As artificial intelligence becomes more ingrained in our everyday lives, it is no surprise that it has led to Nobel Prize-winning innovations. This year, the Nobel Prize in Chemistry was jointly awarded to David Baker, Demis Hassabis, and John Jumper for their contributions to computational protein structure development.
Jumper is a Vanderbilt alumnus who graduated from the College of Arts and Science in 2007 with a degree in physics and mathematics. He later earned his PhD in theoretical chemistry from the University of Chicago in 2017. In 2020, he worked with Hassabis to release AlphaFold2, a machine learning algorithm that can predict the structure of proteins. Under DeepMind, a startup founded by Hassabis, AlphaFold has predicted the structures of over 200 million proteins. To understand why this project is worth a Nobel Prize, it is first important to realize how important proteins are to medical and biological research.
How does AlphaFold work?
AlphaFold uses software called deep-learning neural networks, which recognize patterns in data. The model was first trained on experimentally determined proteins from the Protein Data Bank. When given an amino acid sequence, it estimates the positions of each monomer based on its relation to the other monomers. Certain sequences of amino acids create common structures in different proteins, and by identifying these patterns, AlphaFold can create an entire protein.
Proteins are crucial to every biological process in the body. Because of this, drug discovery relies on understanding protein structure. The first step in the research process is identifying the target. Many drugs bind to proteins to create the conformational changes that are responsible for healing the human body. To design a drug that targets a specific protein, scientists must know the structure of the binding sites.
Until AlphaFold, the primary ways of discovering protein structure were X-ray crystallography and cryo-electron microscopy, a process where electron beams are fired at a protein to determine its shape. While these methods can provide reliable information, they are expensive and time-consuming. AlphaFold takes another approach by using artificial intelligence and machine learning to make predictions about the structure. In 2018, Google DeepMind released AlphaFold1, which placed first in a worldwide protein structure prediction experiment called the Critical Assessment of Techniques for Protein Structure Prediction (CASP). Even though it was the best software at the time, Jumper and his team believed that improvements needed to be made before it could be used in place of other methods. AlphaFold2 was released in 2020 and was again the top contender at CASP. This new version of AlphaFold was now a valid option for predicting protein structure, revolutionizing biological research.
Applications of AlphaFold
Despite there being no other comparable software, AlphaFold still does not compare to the experimental strategies for deriving protein structures. X-ray crystallography and cryo-electron microscopy can create protein structures with complete accuracy, while the precision of AlphaFold prediction varies based on the proteins that have already been discovered. This limits how well the model can predict structures with uncommon amino acid sequences. The database does provide the model’s confidence with specific sequences, but it may take a better software to replace the traditional ways of finding protein structures.
This does not mean that AlphaFold is not usable — in fact, it is already being used in drug discovery laboratories. For instance, researchers developing a malaria vaccine have used AlphaFold to predict the structure of a protein called Pfs48/45. If antigens are created to block this protein, mosquitoes cannot transfer the parasite that causes malaria. The challenge is determining the binding sites on Pfs48/45, but AlphaFold solves this problem. Even though the program makes errors, combining it with cryo-electron microscopy gives promising results. The current model of the vaccine created using AlphaFold has entered the clinical trial phase.
Another application of AlphaFold is the identification of proteins involved in rare diseases. For example, alsin is a protein that is involved in rare motor neuron diseases such as juvenile primary lateral sclerosis and amyotrophic lateral sclerosis. AlphaFold is a method for initiating studies for rare diseases like these that do not receive as much funding as widely researched diseases.
Jumper and his team are not done with improving their model; AlphaFold3 was announced on May 8, 2024. Overall, AlphaFold has changed the course of drug development and shown how artificial intelligence can be used to solve previously unattainable problems.
Cover image taken from AlphaFold Protein Structure Database
References
Callaway, E. (2022). What’s next for AlphaFold and the AI protein-folding revolution. Nature, 604(7905), 234–238. https://doi.org/10.1038/d41586-022-00997-5
Doster, S. (2023, August 8). John Jumper, developer of AlphaFold, to present an Apex Lecture on August 30. Vanderbilt University. https://medschool.vanderbilt.edu/basic-sciences/2023/08/08/john-jumper-developer-of-alphafold-to-present-an-apex-lecture-on-august-30/
Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., Bridgland, A., Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B., Nikolov, S., Jain, R., Adler, J., & Back, T. (2021). Highly Accurate Protein Structure Prediction with Alphafold. Nature, 596(7873), 583–589. https://doi.org/10.1038/s41586-021-03819-2
Jumper, J. et al. “Highly accurate protein structure prediction with AlphaFold.” Nature, 596, pages 583–589 (2021). DOI: 10.1038/s41586-021-03819-2
Ko, K.-T., Lennartz, F., Mekhaiel, D., Guloglu, B., Marini, A., Deuker, D. J., Long, C. A., Jore, M. M., Miura, K., Biswas, S., & Higgins, M. K. (2022, September 24). Structure of the malaria vaccine candidate PFS48/45 and its recognition by transmission blocking antibodies. Nature News. https://www.nature.com/articles/s41467-022-33379-6#Sec8
Sebastiano, M. R., Ermondi, G., Hadano, S., & Caron, G. (2021, December 25). AI-based protein structure databases have the potential to accelerate rare diseases research: AlphaFoldDB and the case of IAHSP/Alsin. Drug Discovery Today. https://www.sciencedirect.com/science/article/pii/S1359644621005705
Varadi, M. et al. “AlphaFold Protein Structure Database in 2024: Providing structure coverage for over 214 million protein sequences.” Nucleic Acids Research, gkad1011 (2023). DOI: 10.1093/nar/gkad1011
Varadi, M. et al. “AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models.” Nucleic Acids Research, 50(D1), pages D439–D444 (2021). DOI: 10.1093/nar/gkab1061