12 Comments
User's avatar
Cole R's avatar

The preclinical discovery engine is ramping up but the bottleneck has been and will continue to be clinical development and trials. There simply aren't enough patients to gain enough power to meet the needs of these trials.

ljubomir's avatar

Perhaps AI can help us find more patients for the trials

But I think a key problem is time. Clinical trials take a long time. Unless we somehow develop "virtual patients" or other in-silico substitutes for RCT (a long shot, I'd say), the time will remain the fundamental obstacle to major, dramatic progress in drug development. The article concedes that ("Non-verifiable domain with long feedback loops") but it's a bit buried

Bernard Louvat's avatar

Great piece. One addition from the commercial side: the "annuity" only pays while patients stay on therapy, and on GLP-1s most don't. A JAMA Network Open study of 125,474 US adults found 64.8% of those without type 2 diabetes stopped within a year. AI that keeps patients on the drug through the hard weeks grows the same pie as AI that finds the next one. It's the part of the loop we're building at TeraHealth https://tera.health/ with TeraRx. Do you see post-approval AI, like adherence and patient support, as part of the bio x AI thesis or as a separate market?

Cell FI's avatar

Lilly's GLP-1 portfolio doing close to $17B in a single quarter, nearly Anthropic's revenue, makes the comparison land. But with roughly half of big pharma revenue coming from the top 3 assets, does AI really make pharma more like software, or just make the same power law cheaper to chase?

Tiffany Tian's avatar

The conclusion rests on the assumption that AI can improve the success rate of drug development—which is the premise for a revaluation of drug discovery platforms. However, there are currently only weak signals supporting this assumption.

Ollie Darby's avatar

the hits business then annuity business framing is such a useful way to put it, and it's exactly why I keep going back and forth on Isomorphic reportedly talking about a $50b valuation before it's put a drug into a human. If AI really does lift the odds of getting through trials then that number starts to look a lot less mad, but until there's one actual approval it still feels like a bet on the first half of your argument and not the second

Ben Versh's avatar

Great piece. When describing industry, it is rather rosy and consideration should be give to:

“patent cliffs” - there is no guarantee a company will discover or purchase its next big drug, but there is guarantee that drugs go off patent and generics come. Companies like Pfizer have been in a spin since loss of Lipitor.

Reliance on US for revenue - industry is overly reliant on US for profits vs other industries. Pressure from mid-terms to take action will be real headwind to address.

Speed of innovation in Oncology - fields like oncology have much shorter cycles of “annuity” as innovation is happening so rapidly in that area.

JCraddock | Redacted Science's avatar

You wouldn't need half those drugs if they didn't redact science.

Travis Wright's avatar

A hits business that turns into an annuity is basically the app store model with a 13 year review queue. The part I'd watch is your wet lab point: if AI makes drug candidates abundant, the bottleneck moves to CRO capacity and clinical validation, so the picks and shovels winner might be whoever owns the robotic labs and the proprietary data loop, not the prettiest model. Curious whether you think full stack AI native biotechs can actually outrun big pharma's habit of just buying two thirds of its pipeline.

Jordan Vale's avatar

Your GPT-3 comparison is the part I keep thinking about. If Bio x AI is only at the GPT-3 stage, what is the one bottleneck that has to break before it reaches the next level?

Jojo's avatar

The more I read this SS and a variety of other writings, the more it is hammered into my conscious how much EVERYTHING in our current world is built around making and accumulating money. This is disturbing and the root of so many of our problems.

I look forward to the [hopefully not distant] time when AI and its robot workers do all the work that humans now do and money has become obsolete, having no reason to exist in a then in a post-scarcity world of super abundance.

Dr. Dereck Mush, MD, MBA's avatar

The number behind Eroom's law is starker than the name suggests. Scannell's 2012 paper found new drugs approved per billion R&D dollars halved roughly every nine years from 1950, about an 80-fold fall after inflation, over the same decades when screening, chemistry and computing all improved. Better tools have failed to bend that curve before. For the AI-native drug companies, approvals per dollar is the scoreboard that will settle it, more than speed to a candidate.