The AI boom has created a new kind of pharma company: a solutions provider that partners with outside companies that want to leverage its tech — while also working in drug development and maintaining its own pipeline of assets.
Among this emerging and growing group of companies, Exscientia has become a standout. Formed in 2012 from a spinoff from the University of Dundee in Scotland and based in the UK, Exscientia has since stuck a claim on a few key milestones in the industry, such as bringing the first AI-designed drug into clinical trials.*
Although Exscientia has its own candidates in development, its major partnerships have fed the buzz around the company. Earlier this year, the company announced that it is expanding an ongoing collaboration agreement with Sanofi to include 15 new molecules. The deal, with its potential value of $5.2 billion, represents the deepest investment by a big pharma company in AI to date.
But Exscientia has plenty of other milestones it wants to conquer with AI in the years ahead.
In a sit-down interview with both Sanofi’s chief scientist, Frank Nestle (read the Q&A here), and Exscientia’s CEO, Andrew Hopkins, the pair discussed why they locked in the major collaboration. Here, Hopkins also discusses gaps in AI development that still need to be filled, and how Exscientia is working to expand its AI reach through the entire lifecycle of drug creation.
PharmaVoice: How did Exscientia’s collaboration with Sanofi come about?
Andrew Hopkins: Sanofi’s confidence in us has been based on evidence we’ve already generated. For example, from our relationship with Bristol Myers Squibb, which started over two years ago, we’ve brought two molecules into the clinic. And BMS has licensed a molecule from that relationship and gave an additional endorsement by expanding the collaboration. All together, we’ve been the first company to bring AI-created molecules to the clinic three times, including one from our own pipeline.
We’ve also tested the drug optimization process. It can often take up to 5,000 novel molecules in order to be made and tested [to find one fit for clinical trials]. In our projects where we’ve identified a candidate in the clinics, we were able to do that from about 150 to 400 molecules. This is close to a 20-times improvement in efficiency when we compare it to conventional approaches to drug design. So, we have a repeatable, evidence-based approach, which gives us confidence as we get ready to scale up this approach.
Do those three candidates in clinical trials represent Exscientia’s biggest wins?
There are actually a few major things we’re proud of: Those three clinical candidates, which were AI firsts.
The other major reported AI first was in October, when we published the results of our EXALT-1 clinical trial. This was the first time an AI technology has been shown to improve outcomes in a clinical study in oncology. We had a 30% [improvement over] drugs selected by a clinician compared to drugs treatments selected by our single cell, deep learning analysis approaches. This is an incredible advancement from which our translational models are based on.
We believe these are the key AI firsts in biotech, and as we scale the pipeline internally and with partners we can provide more evidence about the benefits AI can have in drug discovery.
Why is the collaboration with Sanofi significant for Exscientia?
There are two things that are so exciting about this collaboration with Sanofi. One is scale — we’re doing 15 projects over the next few years.
But this is not just about drug design and discovery. It’s also about how we can use AI to selectively prioritize the drug targets we want to work on together, and how we can use AI to select the patients we want in clinical trials. With precision medicine in particular, we’re working on patient-centric tissue models where we actually take tissue samples from oncology patients [to get clinical insights], and we’re working on developing this approach for pathological diseases. And we’re using deep learning approaches to analyze [patient responses] on the single cell level.
So, it’s this application of AI across the entire dug creation pipeline that’s really exciting about this collaboration.
What are the gaps in AI capabilities that still need to be filled?
We want to build out an AI-first approach for the entire drug creation process — from the initiation of the idea to the approval and marketing of the drug. One of the things we [have to do] is continuously accept new challenges, particularly as we develop our own pipeline. As we progress forward and we take on new challenges and move deeper into the clinic, we are excited about using AI for patient selection.
Another area we could improve upon is looking at how to apply AI deeper into the clinic and how we design clinical trials. And then ultimately, how do we go more broadly and expand the types of modalities that could be designed with AI?
How will Exscientia approach making these advancements in the realm of using AI for clinical trials?
We will certainly be looking to make improvements as we face new challenges. And we’re tackling these problems now, even before growing a large clinical pipeline. This year, you'll start to see new announcements about how we grow into clinical innovation.
We’re also working on growing out our physical network to expand the set of relationships we have with hospitals and physicians, particularly in terms of building relationships where we can source additional tissue samples or bio banks and provide additional data on personalized screening back to clinicians. In fact, we've already built a network of over 70 different sites across Europe, and we'll be looking to expand that as we as we go forward. We are actively looking at how to develop our clinical AI strategy.
What kinds of challenges do you face when it comes to working with datasets?
We are in an interesting position because even with the exciting developments of AI in drug discovery, we are in an environment where we have a sparse data problem rather than a big data problem. This is particularly important in drug discovery because when we work with first-in-class targets we often have very little data. A key focus of algorithm development has been: How do you learn from small datasets? This problem has given us an impetus to continuously develop our platforms.
The other thing about data is that not all data is equal. So, understanding what is the most relevant data is why we have such a focus on patient-centric assays. We believe that if we work on the more relevant systems that are derived directly from patients, it gives us more relevant data that we can then translate to a clinic. This is why it was so important for us to show that we can actually have validation on the predictions we make. As I mentioned with the EXALT-1 clinical trial, this validates our approach of putting the patient first. It’s all about the development of the data and that data being relevant to the individual patient’s disease.
Drug discovery is a learning problem. This means the more we do, the more we learn. This is so important to invention, innovation and pioneering in this space. We need to understand the technology so that we can ask the right questions, as well as create the right answers.
*Note: This distinction is a little tricky. According to a recent report in Forbes, Insilico Medicine recently announced that it has become the first company to usher a drug “designed from scratch” using AI into trials. Meanwhile, Exscientia is considered to have brought the first AI drug into trials for an “established protein target.”