Associate at IAG Capital Partners

The pharmaceutical industry is one of the most R&D-intensive industries in the world, with total R&D spending of $139bn in 2022. For a single asset to be brought to market, the process can take 13-15 years and cost $2-3 billion when you include the cost of failed programs. 

Despite numerous innovations and a more than 60% increase in R&D spending over the last decade, the efficiency of the pharmaceutical R&D process has only worsened relative to other industries. For example, a survey by the Congressional Budget Office found that R&D intensity among publicly traded US pharma companies has grown from 12.3% in 2000 to 25.66% in 2019 while, over the same period, the R&D intensity of Semiconductor companies grew from 11.8% to 16.6%, and for the overall S&P, it grew from 2.56% to 3.22%.

To turn this around, a variety of emerging AI technologies are seeking to offer novel, more efficient approaches to the process of drug discovery. In this article, I’ll focus on startups that are applying AI to the early phases of drug discovery and highlight some of their recent milestones and cutting-edge approaches. 

Early Drug Discovery 

The early research and discovery phase of drug development can be broken down into four steps. The first step is the selection and validation of a ‘target’ (a biochemical structure which can be modulated to initiate a physiological response). This is followed by the generation/identification of compounds that are bound to the selected target. These compounds are known as “hits”, and, once their pharmacological capabilities are analyzed, the best compounds are chosen as potential “leads”. Finally, the lead compounds are optimized to increase their binding and therapeutic effectiveness while decreasing their toxicity.  

The Opening for AI drug discovery startups 

One commonality of the four steps outlined above is that a massive amount of data must be analyzed; thus, it is precisely where AI/ML technologies can drive insights. For example, the traditional approach to target selection and validation is quite linear in that it starts with a single target and finds compounds which, preferably, only regulate that specific target. By using AI/ML models, one can leverage exceptionally large datasets to begin investigating more complex relationships such as how multiple biochemical targets work in tandem to drive disease. In December 2021, Recursion Pharmaceuticals partnered with Genentech and Roche to investigate this very problem on a deal providing $150mm upfront to Recursion and potentially up to $12bn in milestone payments. Another recent innovation was announced in October 2022 by a team from CUNY Graduate Center who built a model for predicting human response to novel drug compounds.

"According to a Pitchbook analysis covering 190 companies, over $17.6bn of capital has been invested into AI Drug Discovery startups over the last 10 years."

The other three steps in the early discovery phase of drug development (Hit Generation, Hit-to-Lead Selection, and Lead Optimization), require high throughput screening and constant adjustments of candidate compounds. The application of AI technologies to these steps can enable a virtuous cycle of “design–synthesize–test–analyze” by generating compounds in-silico(I.e., in a digital format), producing them in an automated wet lab environment, running the candidates through assays and/or screening systems, and analyzing the results. Ordaos is one such company that is leveraging this continual learning process to de-novo design and validate a new class of mini-proteins, which they call “miniPRO”, that are more configurable and can fit a wider range of targets than traditional antibodies. Another startup, Deepcure, combines cutting-edge AI with an in-house automated chemistry platform that can perform complex, multi-step synthesis. This capability allows Deepcure to analyze a much broader range of candidates than traditional companies would have access to. By unifying these capabilities, these startups exemplify how AI can generate more data about a candidate and analyze it at a faster pace than traditional drug discovery companies. In turn, this should lead to a higher rate of success in the clinic and a faster time to market. 

Market Environment for AI Drug Discovery Startups 

According to a Pitchbook analysis covering 190 companies, over $17.6 billion of capital has been invested in AI drug discovery startups over the last ten years. Much like the trend in the broader biotech industry, the annual amount of capital invested into AI Drug Discovery startups peaked in 2021 at approximately $4.3bn but has since dropped to ~$1bn so far in 2023. There has also been a slowdown in collaboration agreements between Big Pharma and AI companies, with only 17 through Q3 2023 after seeing 66 agreements in 2022 and 53 agreements in 2021.  

Despite the worsening fundraising environment, the R&D structure of the pharmaceutical industry remains broken. Thus, there continues to be an opportunity for startups to build fundamental technologies that make it faster and cheaper to bring drugs to the market. Moreover, we’ve seen early indications of the value that AI can get to the drug discovery process, with Nimbus Therapeutics signing a $4bn upfront agreement with Takeda on a Tyk-2 inhibitor. It’s an exciting time to be a part of the innovations that will ultimately improve the lives of patients.