UiB AI #1 How AI helped solve Protein Folding - and why it matters

Welcome to the first seminar in the seminar series held by UiB AI. The seminar is open to all UiB colleagues and students. Register now and join us for an interesting talk and delicious lunch in the University Aula.

Registration for this event is now closed. Please contact annette.servan@uib.no if you have questions.

Artificial intelligence (AI) denotes a range of computational approaches that allows computers and programs to perform tasks that would normally require human intelligence. It is well known that AI can enable self-driving cars, phones you can talk to, and help recommend which movies you want to watch next. However, AI can also be used to advance basic science. In this seminar we describe a recent breakthrough in basic science achieved thanks to AI: AlphaFold, developed by Google’s sister company DeepMind provided a solution to one of the greatest challenges in molecular biology.

Proteins are key building blocks and tireless workers in all living cells. Proteins consist of chains of amino acids - the order of which is determined by their corresponding genes encoded in DNA. The human genome encodes ~20,200 different proteins.

Most proteins fold their long chain of amino acids into a specific shape: their three-dimensional structure. Protein structures are a bit like the Rosetta stone in molecular biology; without the knowledge of a protein structure, understanding its function and role in the cell is extremely challenging. Unfortunately, protein structures are notoriously difficult to study and require costly and time-consuming experiments to determine. Knowing the structure of a protein can help understanding how it performs its function - and make it possible to select protein variants with higher efficiency or design drugs modulating or blocking its function. Knowing protein structures thus has major implications for biotechnological and medical innovation.

Following careful experiments with the enzyme ribonuclease in the 1960s, the biochemist Christian B. Anfinsen discovered that the structure (and function) of a protein is determined by its amino acid sequence alone. This insight had tremendous impact on molecular biology at the time and became known as Anfinsen’s dogma. For this, he was awarded the Nobel Prize in Chemistry in 1972. This dogma has been at the basis of decades of research in the protein folding field ever since, attempting to predict protein structures from their amino acid sequence: ”the folding problem”

In the 1960s, determining amino acid sequences was a major task in itself. However, with the invention of efficient methods for sequencing DNA in the 80s and 90s, amino acid sequences of proteins are now deduced instantly from genome sequences using the genetic code.

In this seminar we will describe how AI - and the use of deep learning - has been applied to the “folding problem” - and to a large extent solved it. We summarize some earlier approaches to the problem that give hints about what type of patterns most likely are captured by the networks enabling prediction.

The development of the AlphaFold approach by the DeepMind company illustrates the power of modern AI approaches but also illustrates that in many cases both deep domain knowledge and AI expertise is required to develop good solutions.

We also discuss how the arrival of AlphaFold and related methods revolutionize molecular biology and in particular structural biology and what avenues it will open for new insights and innovations. We discuss possible implications for biotechnological, medical, and synthetic biology projects and their possible impacts.  

---

Nathalie Reuter is Professor at the Department of chemistry and Head of the Computational Biology Unit. Her research group makes use of computational approaches and generally aims at a better understanding of protein structures and dynamics, of relevance for drug discovery or protein engineering.

Inge Jonassen is Professor and Head of the Department of informatics. His research focuses on development and application of methods to discover patterns in molecular biology data.