Startup Collaborates With the Francis Crick Institute To Model COVID-19's Spread Between (and Within) People
Industry Insight Apr 29, 2020 | by Ruairi J Mackenzie, Science Writer for Technology Networks
London-based engineering startup Hadean has ambitious goals. Its distributed computing resources have earned their stock in collaborations with game companies, such as Minecraft developer Mojang and Eve: Online creators CCP Games, where they have been deployed to produce massive, scalable online environments. Hadean’s aims go far beyond gaming, however, and the current COVID-19 crisis represents an opportunity for Hadean’s technology to move up a gear to tackle global problems.
In a collaboration with researchers at the Francis Crick Institute, Hadean hopes to use its simulation solution, the Aether Engine, to better model the spread of COVID-19 both between and within people. Technology Networks spoke to Hadean’s Vice President of Operations, Mimi Keshani, to find out more.
Ruairi Mackenzie (RM): What role do simulations play in terms of tracking the spread of a virus like SARS-CoV-2?
Mimi Keshani (MK): I guess the question is, why would we want to model something? A model is just a simplification of reality because reality is very complex, and to try and work out what's going on, we can only look at abstract pieces of it. I think Imperial's model of disease spread and infection, which Neil Ferguson and his team built up over 15 years, is probably the most detailed model of virus transmission in populations that we have as a global community. The reason that it's such a great model to simulate scenarios for COVID-19 is because it details the outcomes of potential interventions – helping to inform decisions that policymakers can enact. The way their model works is mostly based on a technique called multi-agent simulation and that is the technique that Hadean’s Aether Engine is very well suited to. Multi-agent simulation is basically a way of looking at individuals i.e. in this case, people, and understand how they interact with each other in a given environment. You can create a complex picture of a macro event – like disease spread in a whole community, be it a city, country, or the world.
It’s that ability to look at the individual and understand the non-obvious emergent behaviour in the macro. That’s an interesting problem that Hadean's technology can apply itself to because we've got this distributed simulation engine; something that can scale to very, very large sizes, without the additional complexity that normally comes with that. So why would you want to simulate something? Because it gives us the ability to model different scenarios and make informed decisions.
RM: You mentioned that the Aether Engine is very well placed to advance these simulations, but how are models currently running and how will the application of Hadean's technology improve them?
MK: The kind of modeling that we're talking about with this project is specifically multi-agent simulation. But there are different kinds of models can be run, so you can have simulations based around making approximations of elements that you're interested in. That would be more of a numerical simulation. Where people are taking a more individual focus, looking at how an individual affects a macro, that's the type that Aether Engine is most well-suited to, and when these models hit their limits, then Aether Engine takes over the complexity of scaling.
What we're talking about here with the Crick is a model that is multi-layered and multi-faceted, similar to what Neil Ferguson is doing. I think they're actually open sourcing the code for that very soon in collaboration with Microsoft, so working with them would be something we look to do in the future, to build out that population model to understand the effects of COVID-19 transmission in different scenarios across the country.
The difference is, what we would like to do is layer in a deeper understanding of particular individuals within that environment. So rather than just saying someone is infected or not infected, we want to say someone is infected and they have a higher likelihood of transmission, because we understand their internal biology. We can do that because we've created an additional simulation, that instead of looking at a road network or how people are traveling around the city, we model their lung network and how the virus transmits in their body. The reason we can do that is because we've got the expertise of a great partner at the Crick, Paul Bates and the Biomolecular Modelling Laboratory that he runs.
RM: What kind of data is being fed into that simulation?
MK: In terms of what we're going to do for the internal biology model, we're at the beginning of that project, because there's a number of different ways we can take it. One of the things that happens with pandemics generally is you see the proliferation of open data. Everyone is doing the right thing and releasing their datasets. We saw it with the Ebola epidemic. There is this wealth of data to choose from. But we don't want to create another model that’s just noise. There are so many models flying about at the moment, I'm sure you're aware of many of them.
The question is, which are the most interesting aspects? Which results will provide the highest value, and which will be the most useful elements to drop into that population-level model? We are building on the work we did last year on protein-protein interactions by looking at how these particular receptors in the lungs, called ACE2, work. If you’re familiar with the way that COVID-19 attacks the body, you might have come across it. We could just focus in on just that interaction. But on the whole, it doesn’t look like that interaction is the thing that determines how badly somebody gets the disease or how likely they are to then go on to transmit it or not.
What we’ve been looking at this week is probably more impactful. New data are appearing all the time and we’ve been looking at how the immune system responds, because this is a key determining factor in the likely survival rate of an individual. What’s interesting when you look at the immune system is, you’re effectively looking at a kind of wargame between immune cells and virus particles, which is a problem that lends itself very, very well to multi-agent simulation. We’re working on datasets with the Bates lab at the Crick, their collaborators, and with the open datasets that are appearing almost day by day at this point.
This builds nicely from some of the other things that we’ve been doing with partners at Microsoft. That model that I just described, where we’ve got this immune system wargame, is analogous to how tumor cells interact with cancer interventions, or how thousands of players interact in a Minecraft game. It's a very multifactorial problem.
RM: Will Hadean's model be something that can help us during this current crisis? Or will it prove more informative for later outbreaks?
MK: The thing on my mind is something I alluded to earlier – not creating another model that’s just noise. We want to make sure that the work that we'll put a lot of hard energy into is the thing that's going to create the biggest impact. I think, unfortunately, it looks like COVID-19 is not going to be a short-lived crisis. This virus is going to be around until we're able to develop a vaccine, which is not something that we will have, being realistic, within a year to 18 months at least. Then, that vaccine has to be distributed effectively. What we would love to be able to do is build up a model that is valuable by the time a vaccine is ready, and help support the rollout efforts. That's a conversation that we've already started with a partner who deals with government policymaking. This isn’t just policymaking for vaccine distribution, but reopening of road networks, reopening of schools and other facilities. This is going to be a multi-year problem, unfortunately. So rather than rush something in the short term, we should create something over the next few months with long-lasting value.
RM: Watching the news, we get bombarded with new modeling data on an almost daily basis. Do you have any advice on how to evaluate all this data?
MK: Data that came out of Wuhan recently revised the city’s death toll figures up by almost 50% because of all of the other cases that were not counted during the peak of their crisis. So these daily rolling forecast numbers that we get from the NHS, you actually can't hold much stock by them. The advice I would give is that looking at something on a day-by-day basis, while it feels valuable, is probably not the most informative way of gaining knowledge about what's happening. Looking at data over time is where you're going to see the real insights and value. A model is only as good as the data going into it. And COVID-19 data are emerging all the time. My advice is look at everything with a heavy pinch of salt. Because our understanding of the situation is still developing, and we don't have a full picture. All we know is that we can't trust our data right now.
Mimi Keshani was speaking to Ruairi J Mackenzie, Science Writer for Technology Networks