Skip to main content

Understanding protein structures better

One of the CASP14 targets: a duck hepatitis B virus capsid. © Dr. Tristan Cragnolini

Thursday, 17/12/2020

An interview with CASP14 organizer Maya Topf

HPI & CSSB’s newest group leader, Prof. Maya Topf, was one of the five organizers of the 14th Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP14, 30.11.2020-4.12.2020). At the start of the virtual conference on 30 November, it was announced that the program AlphaFold developed by London-based AI lab DeepMind is capable of determining the shape of many proteins to a level of accuracy comparable to that achieved in a laboratory.

In an interview, Prof. Maya Topf shared her experience of being a part of CASP14 and her thoughts regarding the impact that AlphaFold will have on the scientific community:

What is the CASP experiment? 

CASP is truly a community effort with over 100 groups from around the world participating. The groups are given the amino acid sequences for around 100 proteins and asked to predict the structure of the protein using computer simulations.  These structures have already been determined experimentally but have not yet been published.

How long have you been involved?

This is my second time organizing the CASP experiment. There is an increasing number of cryo-EM targets that are now part of the challenge and as this is my group’s area of expertise I was approached in 2017 to join the organizing committee. This year 7% of the target structures were determined experimentally using single-particle cryo-EM. 

Did this year’s organizers anticipate AlphaFold’s results? 

AlphaFold also “won” CASP13. They were very good so we did expect them to do well this year, but how well they did was a complete surprise.

What impact will AlphaFold have on research community?

The impact is going to be huge if we understand and can repeat what DeepMind has done. We would be able to predict the structure of single chain proteins at the level that is close to experimental but it’s not yet clear whether DeepMind will reveal their methods.  

Hopefully, other groups will understand AlphaFold’s methodology and develop similar programs which will then be made available to the scientific community. This is what happened after CASP13.  The top groups in the field were able to learn and even improve upon the DeepMind method. This time the top groups in the world performed better than AlphaFold did at CASP13.  

How will AlphaFold impact infection research?

If we have a better understanding of a protein’s structure then we have a better understanding of its function. This could lead to improved drug design and novel drug targets.  

This impacts the whole field of structural biology. It’s amazing - now we can take any sequence and determine the protein structure. This will improve our understanding of general disease mechanisms.

Is this the end of experimental structural biology?

No, even if AlphaFold can predict individual protein structures it cannot predict how the proteins interact in large complexes. Proteins don’t function in isolation and we are still far away from understanding how they function in cells. AlphaFold is a step forward – their method is good at predicting individual protein structures but not their dynamics and their interactions with other proteins. Experimental data is still needed.  

How will AlphaFold impact your own research? 

Part of my group’s research involves predicting the structures of large protein complexes and to accomplish this we first need to model the structure of the individual components. If the components’ models are more accurate then the models of the entire complex will ultimately be more accurate.

How did it make you feel to be a part of all this?

It was really exciting! I was waiting expectantly for the news to come out and knew that it would create a big storm in the scientific community. This is a big thing - the protein structure prediction problem has essentially been solved. Scientists have worked on this for many, many years and we thought it would actually take a much longer time to reach this level of accuracy.  The bottom line is that this will push the community forward. 

Overall, it’s been a great experience being involved and not just because of the excitement surrounding AlphaFold. It’s been great preparing the categories for competition, deciding on the targets and working with the assessors. My group has also been an assessor in the last two CASP experiments and we had to determine how well the participants preformed on the cryo-EM targets.  

What will CASP15 be like? 

We hope to have more targets of protein complexes as this is problem that has not yet been solved. For this we need more cryo-EM scientists to give us targets. Potentially there will be interest in predicting protein dynamics – we will tackle even bigger problems.