When a supercomputer models the world, our image of reality becomes clearer

Simulation models can predict phenomena that cannot yet be investigated experimentally. This helps, for example, in preventing disease.

The mysteries of cells, the marvels of the atmosphere, the inner workings of molecules – all of these unfold in a new light when a supercomputer processes data and generates simulation models of reality.

The processing power of computers has grown enormously in recent decades, revolutionising scholarly work in many fields. 

Fresh avenues for research are emerging, and past observations are becoming clearer. Simulation models can also predict phenomena that cannot yet be investigated experimentally. 

“Computational science yields a resolution beyond the precision of microscope imagery. We can see how and why molecules bind together,” says Professor of Biological Physics Ilpo Vattulainen.

Knowing how each atom interacts with others makes it possible to trigger interactions and calculate them for all atoms, hundreds of billions of times. This is scientific experimentation not in a laboratory, but within a computer.

Development of disease

Vattulainen uses computer simulations to model the functioning of molecules in cells. He believes computational science has tremendous potential for promoting health. Preventing disease would be easier if risks could be predicted, for example, from blood samples.

“Our research helps us perceive how a healthy body functions at the cellular and tissue levels and how disease develops. It allows us to understand how problems can be avoided.” 

When Vattulainen established his first research group in 2001 at what was then called the Helsinki University of Technology, he had difficulties finding collaboration partners. At the time, those involved in experimental work did not believe that computational biophysics could serve any purpose. It was not until after four or five years of hard work that Vattulainen began to find potential partners.

Over the past few years, he has co-authored articles with over 30 experimental research groups, including ones focusing on medicine, cell biology, biochemistry and bioimaging.

Speaking the same language

Experimental research generates enormous amounts of data, which computational physicists can use in their simulations.

At the same time, the requirement of close multidisciplinary collaboration is changing the way research is conducted. This is not always easy.

“You need time to learn to speak the same language as researchers from other fields. It takes years just to get on the same page,” says Professor of Computational Aerosol Physics Hanna Vehkamäki.

Vehkamäki and Vattulainen are computational scientists working together with experimental researchers and computer scientists. Creating a valid simulation model requires continuous cooperation with experimental groups, as they provide the key data.

“Research often starts with a question developed together. An experimental group offers its research results for us to test. We then provide our simulation results to them for testing. We interact continuously, digging further into the question,” explains Vattulainen.

No more spheres

Before the advent of computer simulations, physicists relied on pen-and-paper calculations to produce models, which necessitated substantial simplification. Vehkamäki joined the University of Helsinki in 1992 for a summer job in a project exploring whether a computer could solve equations on the speed of particle formation without simplification. The answer was yes. 

“Computing power enables precision. Rather than depicting molecules as spheres, we can now present them in their correct forms.”

A common simulation for Vehkamäki’s group involves establishing how particles that have ended up in the atmosphere interact with each other. Do they form new particles, or are they condensed on the surface of others?

Current measuring devices are not sufficiently sensitive to answer such questions. The concentration of some substances is too low for detection. Simulation modelling has revealed the inaccuracy of many predictions: substances do not behave as expected.

“The details of a molecule make all the difference.”

What’s in store?

Sulphur and nitrogen compounds are decreasing, as regulation has reduced emissions from factories and traffic. Consequently, the atmosphere is now increasingly dominated by organic compounds. The rise in temperature increases their evaporation, but some of them originate from chemicals people use daily.

“In New York, more compounds in the air now come from fragrances and cleaning products than from traffic,” notes Vehkamäki.

The chemical composition of the atmosphere is changing, and we do not know how it will respond. It cannot be measured, as the change has not happened yet.

“We’d like to predict what will happen and which problems we should focus on next”. 

Vehkamäki heads the virtual laboratory VILMA, currently under development. The aim is to create a virtual model of how particles are formed and analyse laboratory test results in more detail.

“Measurements show just one side of the process. Simulation modelling gives us better insight into what the results tell us and what we should measure next. It helps in crafting new research questions.”

Postal offices for cells

Computational science complements experimental research. Multiple independent approaches to a question would reduce the likelihood of erroneous results and interpretation. 

How then can we be sure that a simulation model corresponds to reality? Results are continuously compared with experimental data, and the model is improved until all results are in alignment. The model can subsequently be used to predict entirely new scenarios.

"If the structure of a protein is known, the predictive power of the model is close to one hundred percent. Then it's just a matter of processing the data and interpreting it correctly,” notes Vattulainen.

Several of his group’s predictions have been confirmed experimentally following technological advances. One successful simulation was associated with cell membrane proteins that receive and pass signals. As Vattulainen was interested in how the messages are transmitted, he decided to model the scenario.

“Cell membrane proteins are like local post offices for cells. A protein selects certain lipids around it as ‘lackeys’, to massage its sides appropriately, so to say. If surrounded by the wrong lipids, the protein goes on strike. The mail only gets delivered if the lipids are the right type.”

Miracle of life

Vattulainen first explored simulations in his doctoral thesis. He examined how the choice of materials in combustion engine vehicles can promote the purification of exhaust gases moving on different surfaces. He eventually transitioned away from material science, however. 

“As a child, I often wondered what’s inside us, what keeps us alive, how nature renews itself, how life goes on. As my doctoral thesis progressed, I realised I might have the capacity to find answers to these questions too.”

At the time, computational biophysics was not taught in Finland, so Vattulainen moved abroad to work as a postdoctoral researcher. On his return, he began to combine computational science with biophysics with the aim of emulating nature with maximum precision.

He has never studied life science as such.

“My teaching now covers the full breadth of biophysics and life science. Yet, the only course in biology I’ve ever completed was in general upper secondary school. As long as you’re motivated, you learn by doing.”

Limits of computing

CSC, the IT Centre for Science, offers services to all researchers in Finland. At its disposal is Lumi, one of the world’s fastest supercomputers. 

“Finland’s a great place for computational research. In many other countries, you’re often not granted the time required for computing, but here that’s not a problem,” says Vattulainen.

Before computing power increased, the problems to be studied had to be selected within the limits of technological capability. Now more complex questions can be addressed. With the arrival of quantum computers, the horizon of possibility is continually expanding. But there is always a limit that cannot be crossed.

“You won’t be able to simulate an entire cell at the atomic level, not even in 25 years. And it’s damn difficult to model water,” says Vattulainen.

Brute force

Vehkamäki notes that if you have a high number of complex compounds, the calculations will get out of hand.

“At first we thought that we could calculate all possible molecular combinations in the atmosphere, but found out we couldn’t brute force our way through the calculations. It would take more than the age of the universe even for a supercomputer.”

The atmosphere has over 100,000 different compounds that react with each other and form new particles. The number of possible molecular combinations is practically infinite. That is why machine learning and AI have been brought into play. Molecules are grouped by their behaviour to reduce the need for computing power.

Although Hanna Vehkamäki praises CSC’s services and says that Finns are lucky to have such great access to computers, not even they are always enough. 

“The more you have, the more you want – that goes for computing power as well. We focus on considering what to calculate precisely and what to simplify, and how to find a way through when computing power is insufficient.”

A high point

Despite the limitations involved, modelling can offer insight into many phenomena. A good example involves the mechanism of the antidepressant Prozac, which Vattulainen investigated with Eero Castrén’s group. 

It turned out that the drug did not control the protein that had been assumed, but affected a different one altogether. The research also revealed that the drug binds poorly to the receptor and is easily detached. This means a substantial amount of the medication must be present in the body for it to remain effective. Clinical observations support the researchers’ finding: Prozac must be taken in significant quantities and for some time to take effect. 

This has been one of the high points of Vattulainen’s career with simulation models.

“When a simulation model can confirm a phenomenon and reveal something new from the data, you may be the first person in the world to see it. It’s the best thing, nothing beats it.” 

The article was published in Yliopisto magazine 9/2024 in Finnish.

Seeking patterns

Professor of Digital Humanities Eetu Mäkelä investigates how high-performance computing can support the human sciences. He himself is a computer scientist who ended up in the field of humanities somewhat by chance, through international collaboration projects.

“I realised that these disciplines work with incredibly interesting and challenging questions and datasets. They are just what I need.”

Mäkelä collaborates with folklorists, historians, media scholars, linguists, language technologists and computer scientists.

Computational science can boost traditional qualitative research by uncovering interesting new documents. Large datasets can also be processed quantitatively, and significant trends be identified.

“I’m interested in whether general patterns can be found in datasets and, if so, how. Material produced by humans is complex. When using computational methods, data must be simplified, but this easily leads to trivial and obvious results.” 

Multidisciplinary collaboration is difficult, as all disciplines have their own notions of how research is conducted and published and what different concepts mean.

“Multidisciplinary collaboration often goes wrong because the participants have no idea how research is conducted in other fields,” Mäkelä points out. 

Everyone should be interested in each other’s perspectives. 

“In addition, reading groups should be organized to familiarize with the basic concepts of each field.”

Sometimes collaboration partners propose overly difficult research questions for computers to answer. Narrowing down the questions requires balancing between what is technically possible and what is interesting from a research standpoint.

“It’s very rare for either partner’s first idea to work. Projects are created and shaped through dialogue,” Mäkelä notes.