Morten Nielsen

Vaccine development requires super-fast computing power

Medicine and medico technology Health and diseases Computer calculations

Calculating large amounts of data is crucial in combating pandemics and in developing individual treatment of patients.

The more valid data is available, the better the opportunities. The faster the data can be calculated, the sooner researchers can start developing preventive methods, treatments, and new vaccines. This requires fast computing power and reliable mathematical models. Morten Nielsen, professor at DTU, develops such models and algorithms.

“We still lack sufficient data to enable computer models to give us the comprehensive biological understanding we need to develop vaccines against, for example, cancer. But they’re coming. With the help of data and algorithms, our models and predictions are now more accurate than lab tests. If there’s a discrepancy, we can say almost with certainty that our result is right and that the errors are in the test,” says Morten Nielsen.

Training models of the immune system

Morten Nielsen’s group is designing mathematical models of the immune system to predict how it will react to a cancer or coronavirus vaccine.

“We want to be able to understand individual immune responses. In other words, why each of us reacts differently to viruses. Some get very sick from coronavirus while others don’t. We want to model this as best we can with our mathematical models. And these models have to be trained,” says Morten Nielsen.

Some molecules in the immune system are very important when it comes to determining how the immune system responds, and these molecules have already been described with large amounts of data.

“What happens is that a molecule takes a fragment from, for example, the SARS genome and shows that fragment to the immune system on the surface of an infected cell. If we can predict which fragment from the SARS genome will be displayed, then we can also begin to understand how the immune system will be able to respond to it,” explains Morten Nielsen.

Algorithms look for patterns

Morten Nielsen’s research groups use machine learning, whereby programs or algorithms look for patterns.

“These algorithms are relatively complex and require both memory and time, because they have to look at millions of data points to learn the patterns. We need many different models and large ensembles to get a complete description of the diversity among all people.”

The researchers can train the models in a few days because they have access to Computerome, which is a national supercomputer with massive computing power located on DTU Risø Campus.

“Normally we train 500-1000 models in parallel. This means you need access to 500-1000 cores. I can reserve them for myself for three days, but only on a supercomputer. If it was my own little HPC cluster and I had to share it with others at the department, it would take much longer.”

Morten Nielsen has high hopes of being able to crack the code for cancer.

“Cancers are characterized by the fact that they change in the genome, that they mutate away from the normal genome. So if you can find the different pieces of a cancer genome that’s displayed on the surface of the molecules I mentioned, then you can use them to develop a vaccine that’s specific to the individual person.”

On the floor below Morten Nielsen’s office, Professor Signe Reker uses the mathematical models and predictions to select peptides that can be recognized by immune cells in the cancer patient. This makes it possible to design a vaccine for the patient which in many cases can cure them or prolong their life. The method is already in clinical use.

“You inject a vaccine that makes immune cells that respond to the specific fragments you’ve found. Then the immune cells will kill the cancer without killing the normal cells like chemotherapy does,” explains Morten Nielsen.

A matter of time

When the research team started the project 15 years ago, they only had 512 data points to train with. Today they have six million, and Morten Nielsen expects that the amount of data points will double in a few years and that the supercomputer can ultimately save resources.

“In the old days, you had to measure everything in the lab with pipettes and so on to get answers, for example to what is immunogenic in SARS or the coronavirus, and there are around 10,000 peptide fragments in that virus. You could start from one end and measure them and then be finished in five years, but then you’d discover that out of the 10,000, perhaps only 50 are actually immunogenic in a given version of the virus. If you can find the 50 by only measuring 100 of them, you’ll get the results and the treatment much faster. We can already do that with our tools.”

The researchers still face a challenge when it comes to understanding the biological interaction patterns, in other words why the T cell will bind to one peptide but not another.

“This is a whole new world that we’ve only just started getting data on. Because without data, we can’t do any of this. But when we get data—and we will over time—I like to think we’ll be able to overcome this challenge by training our models with large data sets, as we’ve done in the past. Of course, the models will be more complex and will require more computer power, but there are no obstacles apart from needing a lot of data. It’s about constantly ‘squeezing the lemon’ and seeing how far we can go, then there’ll be a breakthrough.”

Top photo: Professor Morten Nielsen (Photo by Jesper Scheel)


Computerome is one of Scandinavia’s largest high-performance computing (HPC) clusters dedicated to life science research. Secure and adaptable, the cluster has 20 PB of storage, more than 31,000 cores, and a library of over 3,000 different life science applications.

Computerome is funded by DTU, the University of Copenhagen, and the Danish e-Infrastructure Cooperation (DeiC).

Read more about Computerome