Coin Flipping to End the Pandemic: Don’t Workout your Thumbs, Weight your Coins
Open letter to Pharma: Embrace emergent technologies and jettison methodological orthodoxy
I write this wearing two hats. One, of a troubled New York resident experiencing the calamity from coronavirus, anxiously anticipating any news for a cure or vaccine; the other, of a technology investor helping to build deeply technical, often biotechnology-focused companies.
Recursion Pharma, where I sit on the board, published a recent blinded study assessing the efficacy of over 50 ongoing clinical trials for cures against coronavirus. The results? A near perfect bell curve with the mean being only slightly better than having done nothing at all. Despite the urgency brought upon from a global pandemic and impending economic depression pharma writ large is still blindly shooting from the hip when hunting for cures against disease. The industry’s hand-picked, highest potential drugs entering human trials are still only marginally better than random. The ongoing pandemic is painful proof this is no longer a viable status quo. Pharma has to move away from hoping to land in the thin right tail of gaussian luck, and instead adopt emerging technologies to fundamentally shift the curve and odds in its favor.
If we ask each of the 1.6M residents of Manhattan to flip a coin 10 times, about 1000 of them would land 10 straight heads. Next let’s bet each of those 1000 lucky flippers: $100 they can’t repeat that feat. Only the most brazen and statistically naive would take us on that bet. Getting lucky once has no bearing on getting lucky again. They would be confusing their initial luck with “expert thumbs.”
The top ranks of pharma are going through a similar internal reckoning of ego and culture as automation, increasingly cheap and powerful biochemical toolkits, and machine learning come to bear. The executives and boards that sooner realize strategic superiority in weighting their coins rather than reliance on their “superior” thumbs will usher in the next generation of enduring biotechs. Those that don’t will risk furthering an already unprofitable IRR on R&D, the core vehicle delivering pharma’s future therapies and revenue.
The last five decades of pharma R&D demonstrated phenomenal invention and science: the rise and hegemony of small molecules as therapeutics, the introduction of recombinant DNA, the development of antibody drugs, and most recently, the first on-market gene and cell therapies. But this exists in the context of large and increasing numbers of clinical failures, most succinctly captured in the exponentially increasing cost of bringing a novel therapeutic to market. With self-aware irony those in the industry refer to this trend as “Eroom’s Law”, the syntactic and semantic reversal of Silicon Valley’s increasing power of compute, colloquially dubbed “Moore’s Law.”
I suspect we have captured the low-hanging fruit in biology. Our ability to reason about underlying cause-and-effect is limited by our knowledge of underlying human physiology — a physiology whose complexity is approaching the limits of what is epistemically accessible, let alone reductively tractable.
Pharma has long valorized the lone scientist who in the face of all odds has a “eureka” moment — discovering a new pattern of disease cause-and-effect. There’s industry mythology around the scientists who — against convention and data — make “gut calls” to explore the unknown.
Despite the genius of these scientists, the integral summation of clinical readouts from their reductive hypotheses results in data that’s indecipherable from flipping coins: not much better than random and ultimately, tragically, ineffective. Case-in-point: the dozens of recent clinical failures for Alzheimer’s, all of which tested small variants of a particularly troublesome but scientifically en vogue hypothesis positing a protein fragment, beta-amyloid, as causal for the disease. For shareholders and society this proved to be an enormously costly target — a red herring of immense capital and human cost.
In the past scientists discovered new biology, found new targets, developed new therapies, and ultimately climbed the executive ranks of top pharma. Today these scientists are predominantly the culture-setters and decision-makers at the helm of industry. But for every existing scientist-turned-executive there are troves of peers of equal-not-lesser scientists who happened to make the unlucky bets. In essence these were the majority of Manhattan residents who didn’t flip 10 straight heads. Not because their thumbs were less accurate but because they didn’t flip 10 straight heads.
A decreasing cost of biological experimentation coupled with machine learning to deliver novel hypothesis-free insight is fundamentally changing how life sciences are practiced. For the first time in history our methodological tool kit is expanding beyond “train and breed the best thumbs” to “let’s weight these coins in our favor.” Emergent technologies from CRISPR, to automation and deep neural networks push the burden of discovery from “lone scientist” to machine.
Properly implemented this fundamentally removes human bias and uncovers patterns and truths otherwise indecipherable to any individual or group of scientists. We no longer need to hope we’re searching in the right cohort of coin flippers, but can now change the nature of what we’re flipping.
Global pharma has an immense opportunity, though at a steep cultural and methodological cost. Now is exactly the time for such a major shift. Companies that emerge will no longer solely prioritize individual scientific genius hunting for eureka, but rather will instill cultures of software fundamentals taken as given in big tech: data interoperability and transparency — shockingly rare in pharma today, digitized experiment tracking, statistics steeped with appropriate data to guide all decisions, and machine learning to discover novel biology and chemistry. As an analogy Google doesn’t hire physicists to themselves unearth the structure of the web; instead the search engine hires brilliant minds to teach machines to uncover the structure of an otherwise intractably complex web.
Executives and boards of leading pharma: you have a chance to acknowledge your great but increasingly insufficient thumbs and instead weight your coins forever in your favor. As shareholders and societal stakeholders, this can no longer wait.
