In the 2002 Tom Cruise science-fiction film Minority Report—loosely based on the eponymous 1956 short story by Philip K. Dick—a specialized police department apprehends criminals based on foreknowledge. In the year 2054, the “Precrime” police program uses three clairvoyant humans (“Precogs”) to visualize an impending homicide, providing officers with data to determine the crime’s location and apprehend the perpetrator before the crime occurs. Theatrics and philosophical arguments aside, the ability to predict threats to individuals and humanity has captured our imagination. And with COVID-19 not even in the rearview mirror, viral pandemics seem like a perfect place to start.
Last month, Flagship Pioneering announced the launch of Apriori Bio, a health security company providing variant-proof protection against rapidly evolving viruses. Founded in 2020, Apriori has developed Octavia—a unique biology-informed AI platform that can infer the full landscape of potential variants for a given virus. Using Octavia, Apriori can define the antibody repertoire that most broadly protects against current and future virus variants, enabling the development of variant-proof vaccines and antibody drugs. Flagship’s initial commitment to Apriori Bio is $50 million.
Octavia generates maps of existing and potential viral variants based on their ability to bind human cells and evade antibodies. Using these maps, the company aims to define the combination of antibodies (antibody repertoire) that would best protect people against current and future threats and fill gaps in immune defenses with new medicines – specifically, vaccines and antibodies. Unlike fully computational approaches, Apriori’s platform is designed from the bottom-up to seamlessly integrate experimental insights and artificial intelligence. Octavia relies on several enabling technologies like protein display systems, deep mutational scanning, single-cell sorting, and deep sequencing.
GEN Edge got a peek into the newly launched company from Lovisa Afzelius, PhD, Origination Partner at Flagship Pioneering and the co-founder and CEO of Apriori Bio, who provided insight into the inner workings of Octavia and the company’s business vision.
GEN Edge: Was Apriori Bio launched as a reaction to the COVID-19 pandemic?
Afzelius: We actually started Apriori before the pandemic. We looked back at the ten last years or so, and we saw that every year or two, there was an outbreak that could just as well lead to a pandemic. While it was predicted that there would be a pandemic, we were utterly unprepared. And as the pandemic started, we also realized why it’s so hard to be prepared in many aspects to understand how viruses evolved. For example, we took the receptor binding domain of the SARS-CoV-2 and mathematically looked for potential variants. With just seven amino-acid substitutions, we get theoretical numbers of 1021, which is more than the amounts of data ever captured on earth! So, it’s going to be hard to predict precisely the next variant when we think about the massiveness of that space. And it’s tough to make a universal vaccine.
Instead of predicting the following variant, we looked at what antibody repertoire we need to be protected against all future viral variants. What if we could define what that antibody repertoire needs to look like? If we could do that, we could start designing variant-proof vaccines and antibodies that create that antibody repertoire to protect us — not just to today’s but also the future variant threats.
We set off building a platform called Octavia, a biology-informed AI platform and a highly generalizable platform. It’s a way to apply it to all viruses with ultra-dynamic features that can constantly change. That makes it applicable across not just SARS-CoV-2 but also flu, HIV, and other viruses. We have an entire preclinical development team internally. All the data we feed into the Octavia engine is from tons of internal experimental data. That’s core to what we do. We create insightful biological data, which we use to train our AI algorithms. We can constantly expand the space we experimentally cover and, by doing so, the space we can infer using our artificial intelligence.
GEN Edge: How does Octavia work?
Afzelius: We start by looking at the evolutionary or phylogenetic trees of viruses, look back to what has happened in the past and identify anchor strains or antigens that define viral evolution. Then we start building massive libraries around each of these different strains. We create millions of synthetic variants and test how they bind experimentally to antibodies, which could be from a vaccination, after natural infection, or therapeutic antibodies. We use that experimental dataset to train our AI models. What we’ve done uniquely at Apriori is that we’ve trained our models to understand epistasis, which we have not seen others do. So not just taking a single mutation but also being able to model how different mutations interfere with or affect each other. And that is one of our algorithms’ critical differentiating features.
GEN Edge: What is Apriori Bio’s vision for the application of Octavia?
Afzelius: We are looking to apply this platform across ultra-dynamic biology, starting with ultra-dynamic viruses. We’ve mostly worked on SARS-CoV-2, but we will also model other viruses like flu and HIV. Our vision is to continue developing the platform as an engine that can run both forward and reverse. When you run it forward, it’s about creating intelligence around each new sequence as they emerge. When you run it in reverse, it becomes the time machine for which you can start to design those vaccines and improve vaccines and antibodies that will maintain protection for years to come. The ultimate goal is to apply that across a broad spectrum of viral threats.
We’re moving quickly and growing as fast as possible while maintaining that organic or optimal growth where we keep the quality in everything we do. It’s a balance of trying to do it as thoroughly as possible while at the same time running as fast as we can.
What’s exciting for us is the idea of building something generalizable. This platform’s immense promise is thinking about the ultra-dynamic biological spaces and these concepts that happen across many different viral families. And if we can use biological data to model it using artificial intelligence and thereby infer it to spaces that we otherwise cannot capture experimentally because it’s just not possible. They’re constantly unlocking an entirely new way of being able to design medicines for the future. That way, you already have guarded or protected yourself from what is to come.
We are looking for diversity on the antibody and variant sides. That way, we can also start thinking about the optimal cross-reactive antibody net that is needed either in the form of antibodies or vaccines. We want variant-proof vaccines and antibodies. It’s a new paradigm in how we think about pathogen preparedness and immune monitoring across populations. We are looking for granularity and are starting to see what we can do. In addition to being an innovative platform, it’s a novel way of thinking about how you need to be prepared upfront and designed under those prospects.
GEN Edge: What is Apriori Bio’s business model?
Afzelius: The main focus of Apriori is to make improved vaccines and antibodies. So, in that sense, it’s a traditional revenue model. And we are constantly talking to partners, public and private entities, and sharing our data. Then there is another aspect: with our engine, we can put in any sequence and instantly assess how this sequence will fold into a viable protein and bind to the human receptor. Will it escape current vaccines or antibodies elicited from natural infection? How can we use this intelligence to support and guide policy and manufacturing decisions and to understand how different subpopulations react differently? While the main aspect is the more traditional pharma model, there is a much-needed additional component, which is more about providing a service and making this available to people.
GEN Edge: How will Apriori Bio approach the emergence of a previously unknown viral strain or species?
Afzelius: On the one hand, the benefit of Octavia is that it’s a biology-informed AI engine. On the other hand, the downside is that it’s a biology-informed AI engine. So, if we don’t have any data, we don’t have any predictions. But it’s exceptionally seldom throughout history for there to be a new virus that just emerges from the marsh. It is just improbable. If the variations are so far from anything we’ve seen before, we will be at the outskirts of the model’s predicted space and need to generate more data in those areas. At the same time, given that it is a generalizable platform, it is amenable to constantly adding more information and expanding the model space with every new data set we generate. So that’s about how we build a digital framework and a pipeline that constantly continues to train and evolve those models.
If it is one new sequence, of course, what we can do is we can run it through the system, and we can see how well the system predicts and what it predicts the risk factors around it. We don’t want to be in a reactive position. Instead, we want to have identified the optimal antibody combination that most effectively covers the viral threat so that when that new variant comes out, our variant-proof improved vaccines would already have closed off the escape route for such a novel variant.
And we would already be protected towards such a new variant. That’s the ultimate goal: to be no longer reactive. Right now, the virus is in the front seat and driving around with us, and we are on the defense basically. We turn that equation around, and we are the ones that are actually on the offense and have already instant preparedness when there is a new sequence that’s coming out. That’s how we want to think about the totality of pathogen preparedness.