AWS Startups Blog

How AWS Helped PostEra Scale a Global COVID Antiviral Project

Guest post from Matt Robinson, CTO of PostEra: A company applying its latest advances in machine learning to drug discovery.

Early in the pandemic, while the world was largely focusing on other important technologies and therapies to combat COVID, those of us in antiviral development felt that the development of a pill to treat COVID was a necessary plan to pursue. With concerns ranging from public anti-vaccination sentiments to the difficulty in delivering vaccines or antibodies to the Global South, we saw a need for an accessible therapeutic treatment. But we had trouble getting anyone to listen.

Recognizing this, in March 2020, we at PostEra helped launch a worldwide effort to develop novel antivirals, using technologies from AWS, one of the early supporters of this ambitious project. Governmental organizations were slow to fund it, directing their resources elsewhere. AWS staff, however, rolled up their sleeves and helped us rapidly scale the infrastructure and innovative technologies to speed compounds toward clinical development.

Post Era Global Contributors Map

Map of the global contributors to the Moonshot consortium.

The goal was a therapeutic pill that would be cheap, easy to stockpile, deploy, and administer. At times, it seemed like a moonshot.

In fact, that’s its name: the COVID Moonshot, and it has since become the world’s largest open-source drug discovery effort. This initiative was unique by borrowing an open source model from software development. COVID Moonshot accepted input from labs in the UK, Israel, US, China, India, and Ukraine. It unleashed the power of machine learning to speed up the process of molecular design by assessing thousands of crowdsourced ideas from chemists to determine which compounds were the most easily makeable. While chemists were stuck at home around the world, machine learning was generating predictions and simulations that compressed weeks of human endeavor into a matter of hours, prioritizing what needed to be made

Building the infrastructure for a global collaboration

PostEra uses our latest advances in machine learning to improve the way we discover new pharmaceutical drugs. From a drug discovery perspective, the COVID Moonshot required a new type of architecture, to combine the ideas of a completely distributed team amid global lockdown with these latest machine learning advances.

In AWS, we were able to change our infrastructure quickly to address these challenges, with services scaling to analyze the rapidly arriving data. Solutions architects provided additional insights on building an architecture where reacting quickly to project needs would be straightforward. About a week after the decision was made to launch the initiative, we had a new website up and running to serve the hundreds of chemists contributing designs.

Post Era Initial Website

The initial website provides a way for chemists around the world to input ideas to the platform. A graph shows the number of submissions over time.

Machine learning helped scientists predict which molecules would be easy to order or make in a lab. Using Amazon Relational Database Service (Amazon RDS), we were able to quickly spin up a multi-terabyte database of more than 10 billion catalog compounds to search. Additionally Amazon GPU G-class spot instances, which are readily available, as well as services such as Amazon Elastic Container Service (Amazon ECS) and Amazon Elastic Kubernetes Service (Amazon EKS), proved ideal for building the machine learning infrastructure needed for this technology to scale and provide the necessary predictions..

Post Era Initial Fragment Screen

The initial fragment screen performed by collaborators at Diamond Light Source identified promising compounds that nicely fit in the protein.

Eventually, these tools were developed into the PostEra Manifold platform. Manifold runs completely on AWS, with an isolated, reproducible, and customizable architecture. The COVID Moonshot initiative has gone on to receive a more than $10 million grant from the Wellcome Trust to continue developing anti-SARS-CoV-2 therapeutics, work facilitated by how quickly we can iterate on our computational tools with AWS.

Since Moonshot launched, we’ve continued to innovate using the AWS infrastructure stack. Moving our tools over to AWS lambda allows us to handle workloads that seemed almost unimaginable a couple of years ago. Our clients can assess the synthetic accessibility of more than 100,000 molecules in an hour. We continue to integrate further AWS tools, to ensure the utmost security and utility for our clients, who can integrate their own custom data. The infrastructure that powered a distributed global initiative of antiviral drug discovery scientists is now helping to drive biotech and pharma drug discovery projects more broadly.

Meanwhile, the need for antivirals seems more urgent than ever. Scientists warn that COVID is becoming endemic. That means, we’re going to have to live with it. And that won’t be easy, even with the development of lifesaving vaccines and monoclonal antibodies. Storage and delivery difficulties, access issues, and anti-vaccine sentiment continue to impede global vaccination efforts. All the while, the Delta variant and others keep raging through unvaccinated groups, even breaking through vaccination.

Recent results from trials indicate that antiviral pills could provide crucial treatment, while adding a needed tool for variants that evade vaccines and antibodies. COVID Moonshot is still hard at work in continuing this mission, as we advance these molecules further towards human trials. Meanwhile, we at PostEra, using the resources of AWS, continue to refine machine learning technologies, such that the drugs of the future can get to humans more quickly.

Matt Robinson is the Chief Technology Officer at PostEra. As both a software developer and drug discovery scientist, he works to bring the latest advances in machine learning and computing to the biotech industry. He previously did research in biomolecular simulation and machine-learning for drug discovery at Yale University and University of Cambridge, before co-founding PostEra with his graduate school advisor.