The Nobel Prize in Chemistry 2024 will be awarded this year to three researchers for developments related to deciphering and designing protein structures. David Baker will receive half of the prize, while the other half will be shared between Demis Hassabis and John Jumper. Their research has revolutionized our ability to predict the three-dimensional structure of proteins and to design proteins with specific functions, a breakthrough with significant implications for research, medicine, engineering, and other fields.
A complex structure
Proteins are among the most essential building blocks of life on Earth. Our cells are predominantly composed of proteins, which perform a myriad of functions that sustain life. For instance, we couldn't breathe without hemoglobin, the protein that transports oxygen to our cells; we couldn't digest food without digestive enzymes, which are proteins that break down nutrients; and we couldn't defend ourselves against infections without antibodies, which are also proteins.
Moreover, proteins function as the molecular machines that sustain life by building cells and executing vital processes within them.They replicate DNA, assemble cellular components, produce various outputs (often also proteins), transport these products to their destinations, and regulate the concentration of other substances to maintain the cell's internal environment.
Remarkably, nature has achieved an incredible diversity of proteins using a very limited set of building blocks. Just 20 different amino acids make up all proteins. Each protein is a long chain of hundreds or thousands of amino acids. Once linked together like beads on a string, the chain folds into a complex three-dimensional structure. Within this folded bundle, amino acids that were once far apart in the sequence of the chain suddenly come into close proximity and must align in terms of electrical charge, solubility, spatial configuration, and more, both in form and function for that region of the protein. The correct three-dimensional structure is vital for a protein’s function: if a protein folds incorrectly, it typically cannot perform its intended role and is sent for recycling.
Prediction gamens
Deciphering the three-dimensional structures of proteins is a complex task. In the past, the only way to determine the three-dimensional structure of a particular protein was to conduct direct measurements. These measurements were carried out using techniques such as X-ray crystallography, to examine the scattering patterns of rays reflected from protein crystals, or electron microscopy to examine frozen proteins. Determining the structure of a specific protein using these methods could take several years of work, require complex and expensive equipment, and demand significant financial investment—and even then, success was not guaranteed.
For years, it was recognized that if we could predict a protein's three-dimensional structure based solely on its amino acid sequence, we could efficiently and inexpensively determine the structures of many proteins, which would offer valuable insights into their function. This would, in turn, reveal insights into their functions and mechanisms of action. This spurred a competitive race among commercial companies and academic labs to develop methods to achieve this goal, primarily through specialized software. In the 1990s, Google even launched a biennial competition where groups competed to predict the structure of proteins whose spatial structure had not yet been deciphered. Ultimately, these predictions were then compared to experimentally resolved structures to evaluate their accuracy.
Demis Hassabis, born in 1976 in the UK to a Cypriot father and a Singaporean mother, learned to play chess at the age of four and quickly began winning numerous competitions. By age 13, he had achieved the rank of Master. Homeschooled, Hassabis used prize money from chess tournaments to buy his first computer and taught himself programming. By 16, he completed his A-levels and was accepted to the University of Cambridge, though he was asked to wait a year to begin his studies at age 17. During this period, he founded a company that developed video games and continued developing games even after completing his studies, founding several more companies. In 2009, Hassabis earned a Ph.D. in neuroscience from University College London. A year later, alongside two partners, he co-founded the artificial intelligence company DeepMind.
John Jumper, born in 1985 in Arkansas, completed his Ph.D. at the University of Chicago in 2017 after earlier studies in the UK. He joined DeepMind, where he became a leader in developing the AlphaFold system. Originally founded to compete in complex games like chess, DeepMind successfully adapted its technology to predict protein structures.
In the early competitions, the developed software achieved approximately 40% accuracy in predicting protein structures. However, DeepMind’s AlphaFold revolutionized this field. In its first competition in 2018, AlphaFold attained 60% accuracy, and by 2020, it had surpassed 90% accuracy in predicting protein structures.
AlphaFold utilizes the vast database of protein structures accumulated over nearly a century to predict how a protein will fold into its three-dimensional shape. As a first step AlphaFold’s algorithm performs a one-dimensional comparison between the target protein sequence and other known protein sequences. The software then identifies conserved regions - sections of the protein that have changed very little over the course of evolution and are therefore present in similar proteins.
Deciphering the spatial structure of a protein is a crucial step—but not sufficient—for leveraging this knowledge to develop benefits such as new drugs
The conservation of a protein region indicates its critical role in the protein’s function, which explains why it has remained relatively unchanged across generations.This information aids in creating a two-dimensional representation of the protein. Using this data, the software attempts to predict the protein’s three-dimensional structure by comparing it to a database containing the structures of more than 180,000 known proteins. While the initial prediction might not be accurate, the software enhances its accuracy by comparing the prediction to similar structures, allowing it to learn and refine its models to achieve maximum precision.
Deciphering the spatial structure of a protein is a crucial step—but not sufficient—for leveraging this knowledge to develop benefits such as new drugs that bind more effectively to specific target proteins. There remains a gap in our ability to predict the binding of drugs and other molecules to proteins, as well as in predicting interactions between proteins themselves. Additionally, AlphaFold is less successful in predicting the structure of exotic proteins, those whose structures significantly differ from those in our current databases. However, we now know the spatial structure of many proteins, and since there is a certain level of repetition in their structures, the software can accurately decipher the three-dimensional structures of numerous proteins based on this information.
DeepMind has even taken it a step further by publishing the structures of most proteins in the human body. It is important to note that AlphaFold also reports its level of confidence in predicting the three-dimensional structure. Currently, only 58% of the regions whose structures were deciphered using the software are defined as having a high level of confidence, meaning there is a strong likelihood that the predicted structure matches the actual one.
Designing proteins
In 1997, Bassil Dahiyat and Stephen Mayo from the California Institute of Technology succeeded in designing a small protein that had never existed in nature using computational methods and then produced it. The algorithm they used to design the protein was limited to small sequences and could not be applied to the design of all types of proteins, whether short or long. A few years later, Baker and his colleagues developed a method to design and produce much longer proteins. Baker acknowledged the contributions of others, telling the Nobel committee that he "stood on the shoulders of giants."
David Baker was born in 1962 in Seattle and completed his undergraduate studies at Harvard University. In 1989, he earned a Ph.D. in biochemistry from the University of California, Berkeley, and later became a researcher at the University of Washington.
During his work there in the 1990s, Baker developed a software called "Rosetta" for predicting the three-dimensional structure of proteins. Baker and his colleagues realized that they could also use the software in reverse: by inputting a desired three-dimensional protein structure, they could receive the predicted amino acid sequence needed to create the protein.
This marked the emergence of a new field of research: the development of artificial proteins
In 2003, the software enabled them to predict the composition of a protein with a structure the researchers themselves invented—one that had never existed in nature. In the next step, Baker and his team artificially created the protein, and using X-ray scattering measurements, they confirmed that the structure of the artificial protein matched exactly what the software had predicted.
Baker and his colleagues realized they could also use the software in reverse: by inputting a desired three-dimensional protein structure, they could obtain the predicted amino acid sequence required to create it. Proteins developed using the Rosetta software | © Terezia Kovalova/The Royal Swedish Academy of Sciences
This marked the emergence of a new field of research: the development of artificial proteins. This field has enabled the creation of entirely new proteins with predefined properties, such as proteins that bind to opioid molecules, molecular nanomachines, proteins for vaccines, or proteins with special structures designed for developing new materials.
Following the success of AlphaFold, Baker recognized the importance and potential of artificial intelligence in protein design. He and his colleagues integrated an AI model similar to AlphaFold into Rosetta, significantly enhancing its ability to design new proteins. In 2008, Baker received the Sackler Prize from Tel Aviv University for his groundbreaking work in this field.
A week of awards
The Nobel Prize in Chemistry is the third and final science prize to be announced this week. Earlier in the week, it was announced that the 2024 Nobel Prize in Physiology or Medicine would be awarded this year to Gary Ruvkun and Victor Ambros from the United States for the discovery of microRNAs and their role in regulating gene expression. On Tuesday, it was announced that the 2024 Nobel Prize in Physics would be awarded to John Hopfield and Geoffrey Hinton for developing computational tools that simulate the functioning of the nervous system - key advancements that laid the groundwork for artificial intelligence. This means that two out of this year’s three Nobel Prizes are closely related to artificial intelligence.
This year, there were no women among the Nobel laureates in the sciences—a disappointing repetition of 2021, when only men were honored with the prize.
On Thursday, the winners of the Nobel Prize in Literature will be announced, followed by the Norwegian Nobel Committee's announcement of the recipients of the Nobel Peace Prize on Friday in Oslo. The week of announcements will conclude on Monday with the announcement of the winners of the Nobel Memorial Prize in Economic Sciences.
Get the Ynetnews app on your smartphone:
Google Play: https://bit.ly/4eJ37pE
Apple App Store: https://bit.ly/3ZL7iNv