AI Drug Development Bottlenecks, The Workflow Middleware Layer
From a16z:
To understand where AI fits, it helps to decompose drug discovery into its three essential stages:
1.Disease-to-mechanism: Identifying a biological mechanism — a pathway, a target, a molecular interaction — where therapeutic intervention will alter the course of disease in humans.
2.Mechanism-to-drug: Creating a molecular intervention in the right therapeutic modality — a small molecule, antibody, siRNA, gene therapy — that achieves the desired mechanistic effect with acceptable safety and pharmacological properties.
3.Drug-to-patient: Designing a clinical development program that identifies the right patients and assesses the molecule’s effects — beneficial as well as adverse.
The vast majority of AI work in drug discovery has focused on stage 2… We are doing a pretty good job at manufacturing keys, but they are generally for the wrong locks. Even if AI lets us make better keys at an accelerating pace, that won’t improve our ability to identify the right locks. The real bottleneck in making a novel medicine is disease understanding: identifying a biological mechanism whose modification actually changes the course of disease in patients. That, far more than molecular design, is where drug discovery succeeds or fails.
The problem with that bottleneck is that on one hand it requires a great deal of basic research that in many case is not just slow and costly - research topic might be pursued for 20-30 years by multiple scientific groups worldwide -, but it also leads to plenty of dead ends. And on the other hand, discovering a new target, pathway or a biological mechanism doesn’t grant you any legal or commercial protections; you can’t patent it. This makes it commercially unattractive for pharmaceutical companies to pursue in-house, even as it’s table stakes for the long-term viability of the company. This is where universities and research institutes, typically publicly funded, come to rescue, as basic research and the lack of commercial pressure is what allows for the creative experimentation needed to discover something nobody knew even existed. Once the “dirty work" is done, however, it’s race on to develop and protect the therapeutic.
Workflow Middleware
From Ben Thompson:
What Microsoft is building is, for all intents and purposes, model middleware. It’s the harness that holds conversations, long-term memory, repository states, tool definitions and permissions, documentation and database retrieval, and validation, i.e. whether the model worked or not. Microsoft wants to make models a stateless reasoning engine that gets the bare minimum of information it needs as input, generates a response, and remembers nothing. To do this requires building an interface with every possible model the harness might need to support, and then an API layer on top for companies to build on. Middleware!
This absolutely does have a cost in terms of performance. Middleware is a version of the “write once run anywhere” ethos that all-too-often results in lowest common denominator software that pales in performance to truly integrated solutions. Following the Microsoft path means rejecting the cutting edge of AI’s potential.
While this analysis is about Microsoft’s strategy in the tech space, it perfectly describes what’s happening in the healthcare sector in Germany: As the new cross-section Level 1i clinics start treating a mixture of patients in a mixture of workflows, someone - or rather something - will need to manage and orchestrate all of that collaborative work, even as dedicated AI apps increasingly automate and improve individual steps in the clinical workflows. In taking that role, this middleware layer will not just commoditise the software below it, reducing it to an API call and this way making it ever more undifferentiated. But it will also accumulate important clinical usage data, which could be used to further optimise those workflows in alignment to the local resources, structures and needs of the clinic, locking-in the customers.
This is exactly the opposite of what’s going to happen with the specialised service groups, where the opportunity to become the system of record, by means of integrating vertically and collecting all the patient journey data, will translate into an opportunity to own the entire workflow.