The Hadean Platform, a distributed computing platform, streamlines running applications via cloud by removing excessive middleware and helping scale the process – a journey that has taken them from the world of gaming to the modeling a pandemic.
Humanity is obsessed with making and breaking records in absolutely everything, just ask the good people at Guinness. In science, we don’t exactly have a land-speed record for sequencing a genome or characterizing a protein, but we do know how long it takes to discover a therapeutic (typically 1 to 6 years) and get it to market (another decade, with all the tests and trials required). Even then, only about 10% get approved. We have gone from identifying a new virus to having multiple vaccine candidates in clinical testing within 6 months – that is Earth-shattering record breaking. This was unthinkable with SARS in the mid-2000s, but our rapidly advancing technology and researchers dropping everything to work on SARS-CoV-2 have made next-to-impossible a reality.
Scaled Up Computing for Record Breaking Games
A big part of this has been global advancements in computing and processing power, leveraging the power of the cloud. Hadean, a UK-based company, has developed a cloud-native supercomputing platform. Their Hadean Platform, a distributed computing platform, streamlines running applications via cloud by removing excessive middleware and helping scale the process – a journey that has taken them from the world of gaming to the modeling a pandemic.
“Our cardinal application is Aether Engine, a spatial simulation engine, but we also have Mesh, the Big Data framework, and we have Muxer, which is a dynamic content delivery network for high performance workloads,” said Miriam Keshani, VP of Operations at Hadean.
They took Aether Engine to the biggest gaming conference around – the Gaming Developers Conference in San Francisco – and were instantly attracted by massive online gaming and specifically EVE Online. The maker’s demonstrated record-breaking massive scale battles, but that often meant slowing the game down.
“Fast forward to GDC 2019. We were there with the makers of EVE Online, CCP games, and together broke their world record for the most number of players in the single game with 14,000 connected clients, a mixture of human and AI,” says Keshani.
The company has continued to work with CCP Games as well as Microsoft’s Minecraft. In parallel, Hadean also took their Aether Engine to a whole new level – the molecular level.
Spatial Engines, Scale and Biology
Hadean and Dr. Paul Bates at the Francis Crick Institute in London partnered to investigate protein-protein interactions. The group is pioneering a new technique in the field called Cross-Docking, an approach to find the best holo structures among multiple structures available for a target protein.
“The formation of specific protein‐protein interactions is often a key to understanding function. Proteins are flexible molecules and as they interact with each other they change shape / flex in response to each other. These can be major structural changes, or relatively minor movements, but either way a significant challenge in the field is being able to a priori predict the extent of such conformational structure changes and the flexibility of each target,” Bates said.
The method can be used to predict protein binding sites –useful for studying disease and drug design – however, it requires a lot of processing power. This is where the Aether Engine comes in.
“Despite promising results, this method’s additional pre-processing steps (to choose the ‘best’ input structures) make it practically difficult to do at scale,” Bates said.
“Publicly available docking servers rely on shared cloud resources, so a full docking run of all 56 protein pairs investigated [at the Crick Institute] takes weeks to complete. We used Aether Engine to sample tens of thousands of possible conformations for 56 protein pairs, profiled by potential energy, and selected candidates for docking according to features in this energy space,” Bates said. “This sophisticated sampling of inputs using Aether Engine led to a significant reduction in computation time and negated any additional burden brought on by this pre-processing step.”
The research found 10% uplift in quality compared to other approaches, and the Aether Engine significantly reduced bottlenecks around pre-processing and docking, run as a publicly available server.
Modeling Spread of A New Disease
One of the first things we learned about SARS-CoV-2 is how it gains entry to our cells. The Spike protein on the viral envelope binds to Ace2, a receptor on the surface of endothelial cells. By binding Ace2, this effectively acts as a gateway for SARS-CoV-2 to enter our cells and begin replicating, spreading infection throughout the airway.
Buoyed by the success of their first study, the Bates Lab and Hadean renewed their partnership to focus on simulating COVID-19. The Aether Engine’s simulates a model of the lungs, going down over twenty levels, called ‘generations’, at each of which the airway bifurcates.
“In the model, the virus is introduced at the top because we assume it was inhaled. There is a partial computational fluid dynamics element to it, as the virus travels down the airway according to a set diffusion rate. As it travels through the lungs there are elements, also known as agents in this type of model, that the virus agent is able to interact with,” Keshani said.
The model relies on a number of parameters and can be used to measure the effect of treatments on viral replication in the lungs.
“How we tweak these parameters will depend on keeping track of the literature over time. If there is an interaction between these two agents, the virus will invade the cell and ultimately cause it to burst after replicating inside the cell. Some of these agents will go back into the airways and some into the interstitial lung space. But there’s other elements at play, the immune system fights back, here shown by the antibody and T-cell response, and anti-viral drug interventions can be added to the mix,” Keshani said.
It does have its limitations. The model relies on a number of parameters. Simplifying the complexity of the human body and disease interaction by simulating the effect of what is happening, rather than the actual going events.
“It’s not always possible, or even necessary, to go into the level of detail that we’d love to see. It’s about making trade-offs between what’s useful and what’s reality,” Keshani added.
Supercomputing & Future of Drug Discovery
Drug discovery is a long and expensive process. In recent years, artificial intelligence platforms are transforming the process, helping screen drug candidates and shorten the time required to get to clinical trial. Remdesivir was identified by AI platforms scouring existing drugs for potential COVID treatments. But machine and deep learning platforms require a lot of data to train and make better predictions if they are going to break records in drug development outside of a global pandemic. Keshani thinks there is a role for supercomputing here as well.
“If you’re able to create a simplification of a world that can model emergent behavior, which is the kind of simulation Aether Engine is able to scale massively, you can start building a picture of what could happen if you let different scenarios play out,” Keshani said. “And if you run that same simulation with slightly different parameters 100,000 times or 200,000 times, it’s building up a training set.”
