In your opinion, what is the greatest success of the FAH project?
Delemotte: The greatest achievement is surely to involve so many people from all over the world in such a large research project that brings new knowledge about numerous medical and pharmacological systems. The kind of calculations we do in the course of our research allow us to develop new drugs and drugs by showing us visually where drugs bind to their target molecules and how they work. However, such calculations are very complex.
Put simply, the idea behind FAH is to use the unused computing power of many computers around the world to simulate protein dynamics. Figuratively speaking, every simulation is a kind of discoverer. By sending thousands of them in all possible directions, we can tackle tasks that are unsolvable on a single machine.
The biomedical applications of FAH technology are very broad and include diseases such as Alzheimer’s, cancer and the fight against infection. We are currently focusing on better understanding SARS-CoV-2. With the COVID Moonshot project, we are actively involved in the development of antiviral agents and have already identified the first candidates. Animal-based models are being finalized. We are currently working on its applicability in humans. The aim is to develop and produce a patent-free antiviral drug for oral use. This would strengthen protection against new variants of SARS-CoV-2 worldwide.
What is the most important milestone of the past year?
Delemotte: I would say the scientists behind FAH are at “COVID Moonshot” take part. The development of a new active ingredient against Covid-19 relies on contributions from many research teams around the world. FAH has contributed to this with numerous calculations and simulations, and other groups are carrying out laboratory experiments on this basis. There are already promising drug candidates that we hope will go into development and testing.
What’s the biggest challenge in a distributed computing project like this one?
Delemotte: Aside from the technical challenges for which I rely on other members of the FAH consortium, it is arguably this: explaining to the donors of computing time (and the general public) that our research is a lengthy, random endeavor. Much of the progress remains invisible to outsiders and would be difficult to explain, but it is essential for us. Not being able to point to a specific drug and say, “We developed this, and now it saves lives!” makes it difficult to maintain interest for long periods of time.
To your knowledge, did the insights gained by FAH contribute to the development of a Covid-19 vaccine?
Delemotte: No, because FAH does not take part in vaccine research; our methods are not particularly suitable for this. The aim of vaccine development and optimization is to stimulate the body’s own immune system to fight the virus as quickly as possible so that it does not infect cells.
Instead, we concentrate on a single protein in the virus, the target molecule, and use active substances to specifically disrupt its function. These can be small molecules or peptides that form a selective, strong bond with the target. For example, they help to stop the virus from multiplying – this is the case with the attack on the virus protease, i.e. the COVID Moonshot project. Or they are directed against the spike protein of the virus and disrupt its binding to human cells and thus its entry into the cells. With our method, we visualize whether and how strongly drug candidates couple to a specific virus protein and whether they actually prevent its function. This gives us clues as to how we can improve the active ingredient.
Even before the COVID project, FAH used the computing time provided by the participants to research other diseases such as Alzheimer’s, Huntington’s or cancer. Was there any success?
Delemotte: Yes, even some, for example with the Ebola virus. In addition, researchers discovered, for example, new sensitive areas for drugs, so-called “druggable sites” – important in the fight against antibiotic resistance. One problem with drug development is that there is often only a single snapshot of the appearance of a protein. But proteins are not static; they have many moving parts. With the help of the FAH simulations, we are trying to discover these and using the findings to develop new methods of targeting certain proteins with active ingredients.
Antibiotic resistance is a growing medical problem that costs tens of thousands of lives every year and devours an incredible amount of money. One of the most common forms of resistance is that infectious bacteria produce lactamase. This protein breaks down the antibiotic and thus protects the bacteria.
Our simulations of -lactamase showed that normal protein movements open a part and reveal a hidden pocket (cryptic pocket) that was previously unknown. In subsequent experiments, we were able to prove that molecules that are very similar to drugs slide precisely into this pocket, clamp there and thus inhibit -lactamase. We want to use this method in the future on molecules that were previously considered untreatable. [Anmerkung der Redaktion: Eine Open-Access-Version dieser Arbeit finden ist frei einsehbar.]
How do you see AlphaFold, the algorithm for predicting protein structures from Google subsidiary DeepMind? Is it more of a competitor or another tool available to scientists?
Delemotte: With the AlphaFold algorithm, DeepMind has made tremendous progress in predicting protein structures; they have put themselves at the head of the field. I would like to congratulate the team on this! However, AlphaFold does not explain how proteins fold – this is still an unsolved phenomenon. The algorithm also does not answer any of the other questions that are directly or peripherally related to protein folding.
In contrast, FAH originally focused precisely on this question: How do proteins fold in order to adopt their dominant structure. In the meantime, however, the project also includes other related questions in its research, since all of these phenomena are subject to the same physical principles. Synergies between AlphaFold and Folding @ home could move us forward in these areas.
Will there be new projects at Folding @ home?
Delemotte: There are always new projects in development. This is not just about Covid-19, but also other areas such as neurodegenerative disorders, cardiac arrhythmias, other viral diseases or muscular dystrophy. With more computing power, we can simply do more research in all sorts of areas of healthcare and biotechnology.
GPUs now deliver far more computing power than CPUs. Can classic processors still make a meaningful contribution to the FAH simulations?
Delemotte: Definitely! There are two engines for the simulation technology we use: OpenMM runs on GPUs, GROMACS on CPUs. They cover different sub-areas, so both provide helpful data.
Which parameter promises the highest throughput when folding – high FP32 / FP64 computing power, memory bandwidth, cache (size, throughput, latency) or something else?
Delemotte: Modern, fast processors and graphics chips deliver the best performance. From our point of view, it is more important that as many participants as possible donate computing time to us, even if they do not have the optimal hardware. We can distribute our calculations over many computers and then put the results together.