This article in Stat is going viral in the pharma circles at least five of my friends sent it to me for comment - they know I always speak the truth and do not whitewash. But here, I have nothing to add, Brittany Tang did a deep dive and uncovered some facts that have been rotting the industry for a very long time. The best way to avoid stories like this is to publish the methods for Actual programs in top peer-reviewed journals and provide the software you used for the projects on a commercial or even open source basis. We do not compete with either company in the article. The only thing I can add is that when the company is incubated by investor platform to go after a specific trend, they usually go to big pharma top down, make a few big deals that then hire the executives that made the deal. This is the nature of our industry and is a form of corruption. For true merit-based partnerships we need transparent benchmarks and pharma is avoiding these instead preferring to say that they have internal benchmarks but no-one knows what these are. I was very happy to see that Recursion started publishing some benchmarks at least for one of the recent programs in terms of time and molecules synthesized - everyone should be doing this and for more programs. Link in the comments.
Alex Zhavoronkov, your post really hits home. Remember the BCG article "How Successful Are AI-Discovered Drugs in Clinical Trials? A First Analysis and Emerging Lessons" (by Jayatunga, Ayers, Bruens, Jayanth, and Meier, Drug Discovery Today, Vol. 29, Issue 6, 2024), claiming 75 molecules have made it into the clinic, with 67 still in ongoing trials as of 2023. How many of those drugs were truly "AI-developed"? In their list, one drug even made it to launch - but it was a repurposed drug ... I'm sure AI was involved at some point in its selection, but we cannot call it "AI-developed". Insilico Medicine’s commitment to publishing its methods in high-impact journals, sets a standard that many in pharma still need to follow. Only by laying our cards on the table can we truly reverse Eroom’s Law and foster merit-based partnerships. After all, the best innovation is the one that stands up to scrutiny.
Here's a proposal for the AI biotech sector: Come together to craft an "AI R&D productivity manifesto" - with pillars such as transparency, measurability, shared definitions of performance metrics. Then companies can opt in to join it - and if they do, we know that they subscribe to high standards and report metrics uniformly. There are already R&D industry benchmarks (CMR and KMR being two noteworthy ones) that allow pharma to benchmark against each other in a systematic manner (across many dimensions: PTRS, cycle times, costs, etc.). One challenge with AI biotechs is that - overall - they are still early stage and with relatively few datapoints for benchmarking. BUT: If your AI biotech believes they can "beat the market", why would they be afraid to join a transparent benchmark?
I think the problem starts with “AI-developed” having become a badge of honor, a “value” in itself. Good drug is a good drug, no matter how much of a computer time/sophistication vs. lab-bench time was involved in its creation.
This is a classic case of FOMO. Every company out there says they are using AI. so we should say it too. As a result every company with “2 ifs and 1 or commands” in their software became AI companies over night.
Really appreciate being honest about it.
AI can mean a lot of things. I am giving a lecture on this in Phoenix, Arizona on Friday. It is a tool that is useful if it is used correctly in the right context. One of the first things I learned in graduate school in the 1980s was that “artificial intelligence will never replace real intelligence”. What every company and business needs are bright people that can see forward into the future in a way that no computer can imagine.
I thought this might be happening - one can argue it’s efficient R&D as opposed to ‘corruption’ “…when the company is incubated by investor platform to go after a specific trend, they usually go to big pharma top down, make a few big deals that then hire the executives that made the deal”
Someone said, “AI means artificial I, a real I is needed to understand AI.” No one yet is able create something out of nothing. AI needs real intelligent and competent to create a new component.
Move slow and fix things.
https://www.statnews.com/2025/02/10/ai-drug-development-claims-by-biotech-companies-absci-generate-biomedicines-questioned/