Eroom's law is dying. Fast

Eroom’s law, shown above, and coined in 2012, has been a sobering commentary on the rising-cost and diminishing returns in pharmaceutical industry. Unlike Moore's Law, which observed that computing power doubles approximately every two years, Eroom's Law highlights that the cost of developing a new drug roughly doubles every nine years. In effect, it has been a bleak harbinger for an industry struggling with increasing R&D expenditures and slower market arrivals of innovative medicines.

However, the technology landscape is shifting in ways that challenge the applicability of Eroom's Law in today's pharma world.

Here's the good news - Eroom's Law is dying. Fast.

First the disclaimer: It is true human biology remains a tremendously complex poorly-understood system, and the current generation of AI and robotics systems are not yet equipped to create true biology foundation models that can incorporate complex interactions such as cell-cell interactions, or dynamic evolution in tumors and tumor environments, or epigentic interactions with the external environment. The algorithms are imperfect, and associated data is almost non-existent.

In spite of the above, modern techniques are reducing the order of magnitude of the complexity of specific problems to such an extent that they are game changers.

It boils down to a few factors:

1. AI is enabling much faster design-test-design cycles to new drug candidates, in small molecules and in biologics

2. AI + robotics are enabling closed loop design-test-design cycles all the way to optimized drug candidates

A lot has been written about the first two points, but frankly, they don't move the needle enough to move Eroom’s Law.

Remember, in any given year in any given pharma company, only a few compounds make it to clinical trials. This is out of dozens of candidates. Increasing the number of candidates doesn't change anything.

For that we need:

3. AI + robotics + Organoids are enabling not just lead optimization, but potentially reducing failure rates in Phase 2 and Phase 3 trials down from >70% to ~25%

4. AI-driven clinical trial design combined with patient-specific personalization is further reducing trial failures

5. Open-source collaboration and data sharing allowing selection of optimal patient populations

I’ll be talking about each of these in future articles. But here are some take homes:

·????? Organoids based on human tissue both augment and provide a fundamentally superior biological test-bed than mice.

·????? New statistical methods such as Bayesian approaches have already improved trial designs from around 2015 onwards – modern machine learning approaches are pushing that even further.

·????? The advent of personalized medicine has the potential to solve the “better than the Beatles” problem, albeit accompanied with a disruption on how pharma and insurance companies manage pricing and delivery

·????? Many NIH grants now require open publishing of raw data, such as genetic information. Open-source collaboration and clinical trial transparency has led to a much greater degree of transparency and reduction in redundant research performed across pharmaceutical companies.

As we said at the start - here's the good news - Eroom's Law is dying. Fast.

Let’s make it happen.

?

?

?

?

?