It was great to be back in person at the 2022 BIO International Convention this year. It came as no surprise to us at Recursion that artificial intelligence (AI) and machine learning (ML) were hot topics at the conference, with more than half a dozen sessions exploring how these technologies are driving digital transformation in biopharma.
We heard from leaders who are applying AI at various stages across the drug discovery and development process. While the approaches differ, the overarching goal is often the same: bring better medicines to patients with less failure. I was particularly excited to see the variety of perspectives represented - not only from AI-powered companies like Recursion, but also large pharma companies, academics and investors.
There were several themes that came across loud and clear from the discussions. As a company that has been using AI and ML to build maps of human biology for nearly a decade, we’re excited to see these topics go mainstream in the conversations at BIO and beyond.
In the opening remarks of one panel, Colin Hill, Chairman and CEO of GNS Healthcare, referenced the high failure rate in our industry, saying, “At the core of this issue is the complexity of human biology. After decades of molecular biology research, we’re lucky if we know 5% of the circuitry of human disease.” He then acknowledged that advances in biological tools like genomics and phenomics, combined with advances in computing power and AI, are allowing us to think differently about this problem.
It’s energizing to see the range of AI applications to improve our understanding of biology and the bridge to novel pharmacology. Some companies are analyzing real-world data to create “virtual patients” to help inform selection for clinical trials, while others are using AI to optimize the design of new biologics. It was also encouraging to hear broad consensus among speakers that efforts to elucidate biological insights – no matter the path in doing so – will help us find new ways to target previously intractable diseases.
Without fail, every AI-focused discussion led back to the dataset. What qualifies “good” data? How big does the dataset need to be? How do we structure disparate data to optimize and properly train our AI algorithms?
The answers are, of course, dependent on the problem. Clearly defining the scientific question at the outset is critical to understanding what kind of data is needed and how to approach collecting and organizing that data. It was fascinating to hear the variety of methods organizations are employing to build their datasets. Some are creating technologies to collect and analyze unstructured data, such as natural language processing of electronic health records. Others are tapping into public datasets, such as the Cancer Genome Atlas or UK Biobank, and finding ways to mine those data for specific applications.
Data is something we’ve focused on from the beginning at Recursion. It’s the reason we created our own dataset in-house, with the scale, quality and interoperability needed to power our machine learning algorithms and build maps of complex human biology. Today, our dataset is more than 14 petabytes. It continues to grow as we perform millions of experiments every week.
Regardless of the approach, harnessing data strategically is one of the most important factors in creating long-term value. Even the most sophisticated deep learning algorithms are next to useless if not matched with a thoughtfully constructed dataset.
One of the most exciting aspects of the field of AI-enabled drug discovery is the rapid growth in the number of AI-discovered investigational drugs and the pace at which they are advancing. As reported in a Nature article in February, the number of preclinical and clinical assets being advanced by AI-native companies has tripled since 2018. Contrasted with the declining efficiency of all pharmaceutical R&D in recent years, it’s no surprise that AI-enabled discovery has proliferated.
We’re proud to be part of this wave at Recursion. In the first half of this year, we initiated two separate Phase 2 trials for our most advanced drug candidates and are on track to initiate a third later this year. We’re also condensing the time it takes to translate insights from our maps of biology into potential new medicines. Programs in our pre-clinical oncology portfolio, for example, have advanced from initial hit to in vivo studies in a matter of months.
For smaller biotechs, adding the expertise of a larger, more experienced company can be a catalyst for accelerating the technology and opening the aperture to its full potential. In many ways, the intangible aspects of a partnership – a shared commitment to the science, alignment on core values and a strong working relationship among the experts – are as important as the deal economics. These are the factors that are going to support long-term success and allow both parties to learn and grow together, especially as the technology evolves over time. At Recursion, we’re fortunate to have partners who share our long-term vision.
All in all, it was encouraging and energizing to see AI front and center at the 2022 BIO Convention. If you’re interested in learning more about how Recursion is leveraging these technologies to generate novel insights and discover potential new therapeutics, check out our website and follow us on Twitter and LinkedIn.