Bengaluru: The 2024 Nobel Prize in Chemistry has been jointly awarded to three scientists, who also happen to be a chess master, a theoretical physicist, and a biologist, for their work on protein structures. The prize was shared by David Baker for his work on synthesising novel proteins, and the other half jointly awarded to Demis Hassabis and John M. Jumper for their work on using machine learning to predict the structures of all known proteins.
In keeping with the theme of the Physics Nobel, the Nobel Prize in Chemistry too recognizes work done through computation and modelling with neural networks.
All three recipients developed software and neural networks from other fields that they subsequently introduced into biochemistry to solve for protein structures.
Baker’s work is significant for creating novel proteins, especially those that perform new functions. Hassabis and Jumper utilised artificial intelligence to predict the structures of hundreds of proteins synthesised from their building blocks, amino acids.
Also Read: US biologists win Nobel in Medicine for microRNA discovery. Here’s how it prevents diseases like cancer
The ‘Protein Olympics’
The work they are recognized for goes back to the early 1990s, when the community of biochemists was fascinated with how proteins build themselves up from amino acids, and achieve a stable 3D structure. This process is called folding, and is required for molecular interactions, so that proteins can perform their function in a stable manner.
There are over 50,000 unique types of proteins inside a human cell, made up of long chains of 22 types of amino acids. The order in which they fold themselves and the way they are structured creates unique combinations that result in production of different proteins.
Errors in folding result in loss of function or diseases, and thus understanding the mechanism continues to be a focus of research.
The folding problem deals with understanding how amino acids choose to chain themselves in certain three-dimensional structures to produce the necessary proteins, and also why different proteins perform different functions just through amino acid reordering.
Owing to the very large number of combinations possible, working on the problem saw a massive growth when computers came to be introduced in the 80s and 90s. Methods that previously took days to do by hand were done in seconds, and the origins of what is referred to as AI came about.
Spurred by innovation in computing data processing, biochemists and other researchers started a competition called Critical Assessment of Protein Structure Prediction (CASP) in 1994. Participant researchers from around the globe were granted access to sequences of amino acids in proteins whose structures had just been identified and kept secret.
The objective was to predict the structure of the protein using an algorithm that would work with the component amino acid sequences. Thus, the AI model was to predict protein folding. For 24 years, the competition went on without a breakthrough, until 2018 when Hassabis entered the scene.
But two decades before him, Baker debuted his own computer model, Rosetta.
The reverse Rosetta
Cell biologist David Baker began researching protein structures in 1993 at the University of Washington, and developed a software that could predict protein structures called Rosetta.
He introduced the model at CASP in 1998 for the first time. It performed well, which prompted him to flip the algorithm—they developed a new software that could take an input of a full protein structure’s diagram, and then give an output of amino acid sequences that could fit into it.
Thus, novel proteins could be synthesised using existing amino acids resulting in completely new structures and functions. This process is called protein design.
Baker’s research group created a novel protein structure that doesn’t exist in nature, and fed it to Rosetta. The software proposed as output bits and pieces of amino acid sequences. The team then used bacteria to utilise these amino acids in real life and produced the desired protein. This brand new protein that utilises 93 amino acids is called Top7, and is entirely artificial.
Synthesis of such proteins also is closely tied to material sciences, as proteins have unique functions and can thus enhance properties of nanomaterials.
Rosetta also helped produce many other proteins.
Baker has made the code to the Rosetta software open source, and thus it is utilised by many researchers to work on protein folding.
Also Read: Window to subatomic world—all about ultra-fast light pulses that won 3 scientists Physics Nobel
Physics and mathematics in proteins
Hassabis had been a chess master at the age of 13, was a successful game developer keenly interested in AI and neuroscience. His algorithms made many significant discoveries in neuroscience through the 2000s.
In 2010, he co-founded DeepMind, a company that specialised in building AI models to beat strategic board games and chess. It was acquired by Google in 2014, and subsequently beat the human Go world champion two years later.
In 2018, Hassabis’s model AlphaFold achieved an accuracy of 60 percent at CASP, compared to the previous 40 percent, of the target protein structure.
John Jumper studied physics and mathematics, and was employed in a company that used supercomputers to simulate proteins.
While he was working on his PhD in theoretical physics, he also started to use the model to solve protein challenges. Since the company’s model consumed large amounts of computational resources, he started building his own algorithms and computer models.
In 2017, after DeepMind won the Go championship, the model began working on proteins. Jumper applied to work there, and his work improved the AlphaFold model’s accuracy. AlphaFold2 used large-scale neural networks called transformers that are highly efficient at pattern recognition.
AlphaFold2 achieved almost a 100 percent accuracy at CASP 2020 and the competition came to an end.
The future of protein folding
Proteins are responsible for all kinds of functions, and make up other basic structures of biology like DNA and cells. Their versatile combinations allow proteins to be utilised for various metabolic functions across life.
With computers and supercomputers, understanding the minute but intricate structures of proteins is crucial for many aspects in biology and medicine, such as understanding diseases and disorders.
Complex work into protein structures using neural networks is expected to lead to rapid development in synthesis of custom proteins that can perform therapeutic functions, such as help in drugs and vaccines. Proteins can also find applications in decomposition and applications with plastic, combating antibiotic resistance, and more, described by the Nobel Committee as “applications that are for the greatest benefit of humankind”.
(Edited by Amrtansh Arora)
Also Read: ‘Quantum dots’—2023 chemistry Nobel for tiny particles that make your TV clearer & can aid cancer op