By Peter Tummino, CSO of Nimbus Therapeutics, as part of the From The Trenches feature of LifeSciVC
“Over the next five to 10 years, our goal is to become a company that’s leading the world in personalized medicines, a company that is leading the world in productivity, a company with a value of over $100 billion, a company that has five to 10 products on the market …., a company with the strongest pipeline in the entire industry.”
This quote is from the CEO of a technology-based biotech and the sentiments are not particularly newsworthy or unique for today’s biotech observers. We are constantly reminded how we are in the midst of an artificial intelligence revolution of the drug development process which promises to completely transform how we develop drugs with increases in productivity of an order of magnitude or more. AI-based drug discovery biotechs expect that traditional drug discovery will be replaced with an AI-based paradigm that is faster, less expensive and yields a greater rate of success.
But that initial quote isn’t from 2024, it is from an interview in 2001 with Mark Levin, CEO at Millennium (MIT Technology Review). And the breakthrough technology is not artificial intelligence, it is genomics. In 2001, I was at Millennium Pharmaceuticals, one of the major players in the nascent field of genomics drug discovery, and that time has strong parallels to the changing state of drug development today. In 2001 the leadership at Millennium set out goals of transforming how medicines are developed, driven primarily through the technological advances in genome sequencing. Unfortunately, less than one year after Mark Levin’s interview, Millennium laid off 800 employees (about 1/3 of its staff) and the biotech never approached those measures of success. In fact, the Millennium of 2001 (pre-Takeda acquisition) had one drug approved which was Velcade (bortezomib). The proteosome inhibitor, approved for treatment of multiple myeloma and mantle cell lymphoma, has been an important medicine for patients but it isn’t a product of the genomics revolution.
This isn’t meant to imply that Millennium, or genomics-based biotech broadly, failed to transform modern drug discovery – they did. In fact, drug discovery scientists today would be hard pressed to do their jobs without access to a wealth of well-annotated human genome sequences and their association with specific diseases for target selection through to translational medicine. Mark Levin, Millennium, and genomic biotechs were a huge success at improving how we conduct drug development despite the fact that our industry has not achieved the transformational improvements in productivity we all seek. The reasons for the disconnect between a successful transformative technology and major improvements in drug development effectiveness is a topic for a separate discussion.
The genomic capabilities that were brought online for drug discovery (cheaper & faster next-gen DNA sequencing) did not diminish the need for biologists and translational scientists in target and patient selection, respectively. In fact, it increased the requirement for experienced biologists to use the new genomic information and combine it with mechanistic biology to derive new therapeutic insights. Genomics has identified novel gene products with strong disease associations for drug discovery efforts that would not have been identified otherwise. But that turns out to be insufficient, as we need to understand biochemical and biological function of the gene product in the disease setting. Hence, it is the powerful combination of genomics research with mechanistic biology employed today that drives strategy for development of the next generation of important medicines.
Now fast forward to 2024, and the drug discovery community is fully invested in determining how best to apply the capabilities of AI throughout its work. With striking parallels to 2001, the industry is abuzz with highly optimistic claims of a major transformation, where some AI-driven biotechs are explicitly targeting drug development productivity improvements of >10-fold. The value proposition of artificial intelligence (or perhaps more accurately, machine learning) is clear: quickly learn trends across data sets on a scale too large for human consumption, and enhance or replace human molecular ideation with in silico generative design. Indeed, there are even calls in the industry to diminish the leading role of PhD scientists (and ‘the traditional scientific approach’) from drug discovery and drive drug discovery primarily through artificial intelligence.
We have an opportunity in this moment to learn from the industry’s long history of being seduced by transformative technological advances. The medicinal chemist does more than interpret SAR data and suggest new chemical structures for compound optimization. The increase in scale and capacity of current AI/ML computations does not replace the judgement and discernment of the medicinal chemist, which is based on molecular hypotheses and years of drug discovery experience. Hence, with similarity to the genomics revolution and the need for biologists to derive biological insights, chemically aware AI is best utilized as a powerful tool to improve the decision-making of the medicinal chemist.
At Nimbus, we’ve embraced this latter concept and are deliberate in our intent to empower discovery teams of experienced scientists with AI tools. Those teams have access to predictive models that leverage our theoretical understanding of the physics of molecular interactions through simulation and delineate the underlying trends of large data sets using machine learning. More importantly, we seek to derive insights from both physics- and machine-learning-based predictive modeling such that subsequent ideation can be guided with knowledge-based hypotheses. We’ve adopted neither the traditional med chem/comp chem culture whereby computational chemists provide technical expertise in service of a drug-discovering medicinal chemist, nor an AI-only medicinal chemistry approach developed in biotechs recently that minimizes decision-making by experienced successful drug discoverers. Rather, we are moving towards a hybrid model of computational medicinal chemistry. At Nimbus, we have a 1:1 ratio of computational chemists to medicinal chemists, and we strive for equal decision-making on new compounds from both groups. And we’ve set a high bar for cross-discipline training so that we ensure a computational savviness in our medicinal chemists as well as medicinal chemistry depth to our computational team. The Nimbus experience to date is that this merger of scientific disciplines shows less benefit in speed and much clearer benefit in a wider range of tractable drug discovery targets and enhancement in the quality of small molecule clinical candidates developed.
One final parallel to 2001. Earlier in my career, graduate-level molecular biology departments arose in universities at a time when those technologies were transformative to the science. That has largely fallen out of favor, and the technologies of molecular biology are today essential technical skills for the cell biologist and geneticist. Will artificial intelligence become a set of essential high-powered technologies that greatly enable experienced medicinal chemists? Nimbus scientists are betting on it.
