Introducing Alexander Hauser

How did you get into working with membrane proteins? 

Most receptors have an endogenously produced compound that activates them: for instance, the dopamine receptor is activated by dopamine, the oxytocin receptor is activated by oxytocin. However, there are a lot of receptors that belong to the GPCR family, with a seven transmembrane fold protein structure, that are still “orphan” receptors. They lack a known endogenous activator. In the past, I have employed various data-driven methods to look for gene products that are encoding small peptides that may be activators of some of these orphan receptors. By using machine learning methods, we could learn from known peptide activating receptors. Furthermore, employing various evolutionary conservation methods we could find signals for stretches of genes that are highly conserved in between cleavage sites, which are specific motifs that render a precursor to cleave small peptide hormones into being secreted and then eventually, activating receptors.

How do you use structural biology in your research?

I would not consider myself a structural biologist per se. I have a chemo-infomatic background, from where I transitioned into the data sciences of biology. But, like many other people that might not technically fit in the definition of a structural biologist, I contribute to the field in a different, integrative way.

I am taking these new large-scale AI methods such as AlphaFold and derivatives to the next level by employing them to predict receptor-peptide interactions across evolutionary domains. I received a grant from the Carlsberg Foundation to take this concept further and look for across-species interactions. For instance, potentially, there are secondary metabolites, peptide hormones, that have been secreted from the microbiome and may be activating some of these orphan receptors, which would in part explain why we haven't found anything endogenous yet. We have to look across organisms and genomes, as our body does not only contain our own genes but also those derived from all these species that inhabit our guts and other microbiomes.

Furthermore, there are also lots of peptides produced as part of animal toxins, such as snake venoms or conotoxins, and we are trying to predict which components of these toxins can interact with human receptors and lead to their venomous properties. By employing these machine learning methods as protein-protein or protein-peptide predictions we aim to find previously undiscovered interactions across species. So even though I am not generating new experimental structures, it is a form of large scale, data driven structural biology research.

What role do you think AI and machine learning will have in structural biology from now on?

In the line of what the leaders in the field, such as Demis Hassabis (CEO of DeepMind), are saying, the use of AI/ML is obviously a revolution that opens up many new possibilities that normal X-ray crystallography and Cryo-EM would not be able to do. Yet it’s not there to replace these experimental structural methods entirely. Especially in the drug discovery field, where you really need these higher resolution, specific conformations, specific ligand complexes, it’s there to stay. But I think it’s interdependent: the more experimental structures we have, the better the predictions become and the better the tools that enable these AI-based methods. Also enabling us to do larger complexes with the help of cryo-EM. There are a lot of these huge protein complexs that wouldn’t be possible with either method alone. But they push each other to new frontiers. I think a lot of people have realised that it is very important to encourage integrating machine learning into structural biology. It’s great to see that so many traditionally purely experimental groups are now opening up to these very easy-to-use methods, either through collaboration or by employing these methods through web resources or GitHub protocols, which are often really well documented.

Can you also tell us about your other research pillar in pharmacogenomics?

I just started my group in autumn last year, and the second direction we are taking is framed within pharmacogenomics. We are trying to unravel whether we can explain and predict drug response variability through the impact that mutations in drug targets have both functionally and structurally. For instance, by looking at the genetic variability of GPCRs, which are very prominent drug targets. In order to do that, we are using large scale genomic biobanks to look for mutations of interest and correlating these with clinical phenotypes. I'm heavily collaborating with people who are experts in specific receptors doing e.g in vivo or in vitro pharmacological experiments across the world, but also at SUND, such as with Mette Rosenkilde. I am very keen on integrative approaches, and it is a pillar that combines different disciplines: from genetics to in vitro pharmacology, structure biology and epidemiology. We employ, for instance, structure-based variant effect predictions to inform us whether or not these types of amino acid changing alterations may have impact on stability or on signaling and function of these proteins. One of the ultimate goals is to have a clinical impact or implications in personalising guidelines for how drugs are being prescribed to offer an improved treatment selection. 

Finally, do you have any advice for young PhD or Master's student who want a career in academia?

Reach out to the world, and don't be limited by the people around you on your aisle, on your floor. Don't compare yourself but rather be impressed by or be motivated and inspired by everyone who's out there. I also notice that, in regards to the use of machine learning in particular, more and more PhDs and postdocs can do both: they can take an idea from a paper or a GitHub repository, play around with a tool and then get some predictions, go in the lab and actually validate some of them themselves. I think that's really a strength. I quite dislike this separation between experimental biology and computational biology. Everyone doing experiments these days are required to do statistical analysis. And they shouldn't limit themselves into a corner and say “I don't know much about it”, but rather I would like to push them and elevate them to be able to also do more sophisticated computational work, because that is what they are essentially already doing. They just need to trust themselves that they can actually do it. 

Read more:

To find out more about Alexander’s research see Alexander's research profile:
https://drug.ku.dk/disciplines/translational-pharmacology/pharmacoinformatics/

Topics