We’ve all read (perhaps too often) about the incredible promise of AI, machine learning, and computational processing powers to disrupt the discovery landscape, dream up drugs that no one has ever seen, and propel pharma into the modern age. But as time passes and the field matures, it appears as though AI in Drug Discovery looks more like an evolution than a revolution. A big question we ask ourselves as investors is how to choose the most promising technologies, and what does it take for an AI-drug discovery engine to succeed? It’s an elusive question, but we’ll try to share our perspective:
1. A Great Data Asset
Proprietary, differentiated, and high-quality
With algorithms becoming open-source commodities (see AlphaFold and BioNemo), pure learning and inference capabilities are just not enough. A truly proprietary data asset, on the other hand, can provide a competitive “moat”, differentiation, and even additional revenue options.
The obvious problem is that high-quality, clinically meaningful data is expensive to amass. While some companies – like Recursion AI (NASDAQ: RXRX) with its industrial-level cell imaging capabilities – are heavily investing in their own high-throughput wet labs, this is a luxury few can afford.
Luckily, there are always creative alternatives.
One clever approach is data accumulation as a secondary product from a revenue-generating activity. Take our portfolio company Scipher Medicine: their data asset covering autoimmune diseases has steadily grown through the development and commercialization of a diagnostic test. The now-reimbursed PrismRA blood test helps rheumatologists select the most efficacious therapy for their patients. Consequently, the molecular data derived from the blood samples is used to improve and enrich their proprietary human interactome, which is further leveraged for target and biomarker discovery.
What about tapping into existing proprietary data assets held by academic centers or even other companies? Owkin is embarking on such a project. The French company is building an ambitious, AI-ready data collection through a network of principal investigators and KOLs in what they call a ‘Federated Research Network’. By providing a highly interoperable data platform for the use of research institutions, Owkin is integrating a treasure of healthcare and research data from currently siloed sources. For other examples of differentiated data assets check out Quris AI and Verge Genomics.
Building and maintaining a unique and deep data asset is not a light project. However, if done strategically, it can prove to be the gift that keeps on giving: beyond supporting inferences, simulations, and discovery, it can be leveraged towards forging synergetic partnerships with other players and opening doors to additional products and modalities.
In that context, connecting to initiatives like Nvidia’s (NASDAQ:NVDA) Inception program for start-ups and AION Labs’ venture studio for AI technologies (the latter with multiple global pharma and tech leaders) could help young companies better understand the full panorama of stakeholders, the perception of their offering and data asset, and ultimately balance the big dream versus the reality of the industry.
2. A Diversified Business Model
Where SaaS, licensed or partnered assets add runway and resilience without distracting from the main objective
One of our challenges when evaluating AI-driven discovery companies is the scarcity of mature success stories to use as benchmarks. After some massive IPOs (circa 2021) and much fanfare when AI-designed drugs started entering the clinic – pioneering Exscientia (NASDAQ:EXAI) was the first – there were also some unavoidable early failures, raising skepticism and asking the burning question – can AI truly deliver value for drug discovery?
There’s one company that suggests that it can: the ~$2B market cap Schrodinger (NASDAQ: SDGR). With a diversified business model, Schrodinger has a SaaS offering for molecular and material design as well as a therapeutics pipeline, split between proprietary and partnered programs. With 2023 revenue of >$200M, Schrodinger appears to be not just balancing the inherent risk in discovery, but also embedding itself as a household name in relevant laboratories – and in the process building the network for partnerships and collaborations.
With the line between SaaS and services being quite fine at times, pharma-facing SaaS is not always an ideal option for diversification. But a well-thought-out, diversified business model should serve to add resilience and reduce risk, so we’re watching out for new emerging models.
Most of the companies we are currently seeing are gravitating towards a mix of proprietary and partnered assets as a mode of diversification. Choosing the right partner and establishing a collaboration under the right terms is far from trivial, but when done wisely, partnered programs total much more than their dollar value. In addition to providing a golden opportunity to create valuable connections and learn about the Big Pharma universe, they could also significantly enhance platform validation efforts, leading us to our next point.
