"My research can accelerate development of new vaccines"

Name: Magnus Haraldson Høie
Age: 31
Education: MSc in Engineering, Life Science Engineering and Informatics, Sino-Danish Center
Project research field: Immunoinformatics and Machine Learning
Project period: 2021-2024
Supervisors: Professor Morten Nielsen (principal supervisor), DTU Health Tech, Professor Ole Winther, DTU Compute, and Associate Professor Paolo Marcatili, Novo Nordisk Antibody Design

My PhD project is about ...

… design and optimization of antibodies. Antibodies are a protein-based molecule. They can bind to antigens such as a bacterium or a virus, and consequently help our immune system fight an infection. Like all other proteins, antibodies are made up of long strings of amino acids. The strings are folded together in a three-dimensional structure. All proteins in the world are folded together in their own completely unique way.

In my work, I’ve primarily focused on improving the binding capacity of antibodies. I’ve developed software platforms that use artificial intelligence to predict which amino acid sequences will fold into a specific structure that favours binding. This process is known as ‘inverse folding’. When we predict sequences that are likely to result in a structure that binds more effectively to antigens, we can synthesize (chemically produce, ed.) antibodies with these sequences and test them for improved binding capacities.

The research can contribute to ...

… faster development of new vaccines or improvements of existing vaccines, making them more effective.

During my PhD project, I have myself developed or participated in developing a total of four AI-based tools. I developed one of these tools together with colleagues from Oxford Protein Informatics Group. The tool is called AntiFold, and it can be used to design new antibodies. It’s based on Meta’s large protein language model, which is actively used for protein research. We fine-tuned the model so that it became specialized in antibodies.

Another tool I helped develop can predict the structure of a protein—like DeepMind’s software platform AlphaFold, which was a giant breakthrough in protein research in 2021.

However, our tool focuses on quickly predicting the secondary part of the structure (alpha helixes and beta plates) and where surface-exposed parts of the protein are located. As a result, our tool is 1,500 times faster than AlphaFold and requires less computing power, while still being extremely accurate. We’ve already sold a couple of licences for the tool to some companies.

I get new ideas for solutions when I ...

… talk about my research. I don't talk all that much about what is going well. Instead, I prefer to talk about what challenges me, because formulating your problem is the very first step towards solving it. Sometimes I discover the solution already while I'm talking. Other times, I’ve experienced that you can harvest many new ideas from those you share the problem with—even if they may not be working in the same field as you.

It’s been a great day on the job when ...

… something finally succeeds! Most days you have ideas on how to solve a problem in your research, but they often fail, and you’re kind of stuck with a lot of failed tests. And then you suddenly try out something—perhaps something very simple—and it works.

I take a break from my work when ...

... I play sports. I always make time for sports because it helps me relax. Lately, there hasn’t been that much time because I’m finishing my dissertation, but, otherwise, I’ve been doing a lot of crossfit, cycling, and climbing. I also have a weekly squash game. It gives me a fixed schedule and stability.

I became a PhD researcher at DTU because ...

… I became really interested in machine learning while studying for my Master’s degree at Sino-Danish Center in Beijing. It’s a joint study programme in a university partnership between Denmark and China.

I’m a pharmacist from The Arctic University of Norway, but chose to take my Master’s degree somewhere else. In Beijing, I met the head of my MSc programme, Paolo Marcatili, who was a professor at DTU. When I was to write my Master’s thesis, I contacted him, and he gave me the opportunity to work on a project in his laboratory.

After I graduated, I worked for two years as a research assistant at the University of Copenhagen, but when I saw that a PhD position in Paolo Marcatili’s research group was posted, I applied for the position and got it.

As a new PhD researcher, I was surprised by ...

… the fact that no one holds the absolute truth. In high school, I thought that what was written in the textbooks were more or less definitive truths, but, in reality, they are just the best theories we have right now in a constantly evolving debate.

As a researcher, you suddenly become part of the community that can present the latest theory which can challenge or expand our current knowledge, and you can participate on an equal footing in the debate with your contribution.

The biggest challenge I experience as a PhD researcher is ...

… that you’re the only one who can ensure progress in your project. If you don’t work, nothing happens. There is no progress at all. No breakthrough. No new solutions. It all comes down to you. It’s challenging to keep a high momentum on your own for so long.

You also have to maintain motivation when you face obstacles—perhaps some major challenges in your research or problems with some collaborators.

In the future, I would like to work with ...

… artificial intelligence and especially deep learning, and contribute to our utilization of this technology for a faster development of safe medicine. I’m applying for jobs in the United States. There are many start-ups working in the field of AI-based design of new medicine.