Major breakthroughs in protein research awarded

Proteins are made up of around twenty amino acids in advanced structures in three dimensions. The amino acids are connected in different ways that create long, folded structures that determine the function of the proteins. 

Mapping all the different structures of a protein used to be a real labour of love. But that was before this year's prize winners Demis Hassabis and John Jumper presented an AI model, Alphafold2, in 2020, which could predict all the possible structures a protein could have.

Dramatic improvement

"It is a dramatic improvement. In the beginning, a protein was crystallised in order to identify the position of all the amino acids involved using X-ray crystallography, which was over 60 years ago. Now there is a large database with all the different structures, free for us researchers to use," says Gergely Katona, professor of biochemistry at the University of Gothenburg.

Knowing what the structure of proteins looks like in different cells for different functions opens many doors, including in medicinal chemistry. It becomes easier to influence the function of a protein if you know how the amino acids are organised and where to put your effort. But then the drug has to fit in and be accepted by the protein.

Building your own proteins

This leads to the discovery of third prize winner David Baker. In the early 2000s, he developed a way to build his own proteins using amino acids. Through many observations and his own experience, he was able to design new proteins that can be used in drug production.

"Having access to the correct protein structures is crucial for developing new drugs, or for understanding how different proteins interact and regulate cellular functions, both in healthy cells and in disease states. With the work of David Baker's research group, we can now also design completely new proteins that can act as drugs or be used to measure the presence of chemicals in samples," says Leif Eriksson, Professor of Physical Chemistry at the University of Gothenburg.

Leif Eriksson is a frequent user of AI to find the right molecular structure of the medicine that will affect a particular protein.

"The development of AI-based Alphafold has completely revolutionised structural biology in this respect, making available precise protein structures that would otherwise have taken an enormous amount of time to access, and has opened up completely new research directions. It's a fantastic development where the only limit is really our imagination," says Leif Eriksson.

Much research remains to be done

Gergely Katona is keen to point out that even with these major advances, many puzzles remain to be solved in the world of proteins.

"For example, about a third of all proteins in humans lack a stable structure. These cannot be correctly described by Alphafold. So we need to continue with empirical research in parallel with the AI-based research."  

Research on movements

Image

AI research is also underway at the University of Gothenburg on the movements of proteins and not just their structure. Function often involves changes in structure over time, and the time scales of the processes range from femtoseconds (10^-15 seconds) to years. Consequently, the simulations have to take small steps of femtoseconds to be meaningful and those calculations would, according to some estimates, occupy up to 20 per cent of all high-performance computing resources globally.

"One goal of our team is therefore to develop AI methods to speed up the simulation of these techniques by designing systems that allow us to predict how a protein might move on much longer time scales without explicit simulation. Such a technique could enable the understanding of fundamental processes on time and length scales that are more adapted to biology," says Simon Olsson, Associate Professor at the Department of Computer Science and Engineering at the University of Gothenburg.