How computers could develop a corona vaccine

Jens Meiler, Humboldt Professor at the University of Leipzig, explains the new methods of computer-assisted drug development and their advantages in an interview. The bioinformatician and chemist explains where his research starts directly and how computers can support the search for a vaccine against the SARS-CoV-2 coronavirus, which causes the disease COVID-19. Meiler's research group still works mainly at Vanderbilt University in Nashville, USA. His new institute for drug development is currently being set up at the medical faculty of the University of Leipzig with the aim of establishing local research groups with similar skills.

Prof. Meiler, the aim of your current research is to design a vaccine against the coronavirus on the computer. How would you describe that? Where does your research start?

There are many different ways to make vaccines. I focus on the computational approaches. This is a relatively young field. The vaccines we currently use against other viruses have not yet been designed on the computer. But computer-aided development will play an increasingly important role over the next few years because the methods are constantly improving.

My research starts with the structure of proteins on the virus surface. These spikes, which are always depicted with the corona virus, are proteins that the virus needs to enter the human body. The human immune system, in turn, can produce hundreds of thousands of antibodies that all look different on their surface. Now, if a foreign protein is sighted, some of these antibodies will bind with low affinity. But while the human immune system can make these myriad antibodies, there is usually no antibody perfect enough to neutralize this newly emerged coronavirus. It takes some time for the immune system to optimize the first antibodies, which then bind to these spikes with high affinity and thus prevent the virus from further penetrating human cells. Until then, the virus multiplies and spreads in the body, damages cells and organs, we get sick.

How do you do this on the computer?

My partners at Vanderbilt University's Vaccine Center in Nashville examined the blood of corona-infected patients who are considered cured, i.e. have neutralizing antibodies. They sequenced their hundreds of thousands of antibodies and identified those that bind well to the surface protein of SARS-CoV-2. Now we have to find out which of the approximately 5.000 antibodies also neutralize the virus. Some of these antibodies could be used as therapeutic antibodies as early as summer or fall because they will be available sooner than a vaccine. At the moment we are clarifying the mechanism of action: we predict the structure of these antibodies on the computer and calculate how they bind to the viral surface protein. Sometimes we can also further optimize the antibodies on the computer. Tests are already being carried out with the blood plasma of convalescents. It's the same basic idea - neutralizing antibodies in the blood plasma of those who have recovered neutralize the virus in the patient. Unlike a vaccine, a therapeutic antibody works on acute infection but does not provide lasting protection.

Armed with the knowledge of how neutralizing antibodies bind to the viral surface protein, we can begin designing vaccines on the computer. To do this, we only work with a small part of the viral protein, the part to which the neutralizing antibody binds, the so-called epitope, in order to later focus the human immune system on this point during vaccination. Since viral proteins are very flexible, we incorporate mutations into the computer to stabilize the protein. Since we are working with a small part of the viral protein, we combine the protein with other protein building blocks designed on the computer, which automatically combine to form virus-like particles so that the vaccine to be produced ultimately looks like a real virus. Because only then will our immune system perceive him as a threat and then develop antibodies. These test vaccines are then tested in animal models. This is a very complex process.

How much or how long does a computer have to calculate on it?

That is hard to say. We work with computer clusters. That's around 2.000 commercially available computers that are interconnected and constantly running. Around 35 scientists from my group in Nashville are working on them. In this case it is not a large computer, there are much larger so-called supercomputers. We are not necessarily limited by the computing time here. Rather, the computer model is not perfect. Biology is so complex that it cannot be fully mapped on a computer. Proteins fold and bind to each other, these are energetic processes. Like all processes in the universe, these strive for the lowest energy state. But I can't exactly predict the energy on the computer because these relationships are so complex. Therefore we calculate with an approximation of the energy. In addition, proteins consist of tens of thousands of atoms. There are so many arrangements in space that the computer can't fully search through it. Even the very best, the very largest or the very fastest computer cannot do that.

Our drug development is therefore like a spiral: We go into the computer and we want to design the best possible protein as quickly as possible in a week. Then we have to make the protein and test it because the computer programs aren't perfect. We can then use the results to further optimize the calculations. So it's not like you sit down at the computer and after a run you find the vaccine. Computational drug design alone does not solve the problem. But it can contribute a lot to the solution.

What are the advantages of computer-aided drug development?

The benefits are multi-layered. One advantage is the time, because you can carry out more targeted experiments based on the suggestions made by the computer. You can progress faster because the computer program itself already excludes many possibilities, you no longer have to test everything experimentally. Computer-aided design also has the advantage that you don't find something that works by accident, but thanks to the computer program you have understood the mechanism directly. In the history of drug development, it has often been the case that certain active ingredients were found by chance without knowing exactly how they work. We now have an accurate computer model, and if the drug works, it's because we designed it that way. We then already understand the mechanism, and that is a very big advantage. We then know how the substances work, how they work and can also better assess what side effects could arise.

How do these computer programs work? Are they fed with data and information and are they evolving as a result, so that we can hope for a new vaccine in the near future?

This is where my research starts immediately. We develop these programs, so we come up with better and smarter algorithms to be able to predict the results even more effectively. This links basic research with applied research. Without the algorithms that we have developed over the past five years, we would not be able to design a vaccine so quickly now. But developing and approving a vaccine is a lengthy process. Despite all global efforts, this will probably not be expected before spring 2021.

Source: Leipzig University press release from April 17.04.2020, XNUMX