Speed of learning is a superpower in AI-driven drug discovery.
ProtAI: Faster, smarter experiments that produce drugs that patients receive
This week on Life Sciences Today
Today is a landmark - episode 50 of Life Sciences Today! Less than 9 months in - the show is on fire. If you are TechBio founder, I’d love to host you on the show. Reply or DM me and we’ll make it happen.
My guest this week was Kiril Pevsner, CTO and co-founder of ProtAI.
ProtAI is betting that the real value of AI in drug discovery isn’t bigger models, but faster, smarter experiments that actually produce drugs patients receive.
In this Life Sciences Today episode, I spoke with Kiril, CTO and co-founder of ProtAI, an Israel-based biotech developing drugs for cancer and autoimmune disease. ProtAI’s AMES platform combines structural proteomics with AI, using experimental “geometrical clues” to refine protein structures down to subatomic resolution. That extra precision lets them design small molecules for targets that were previously considered “undruggable.”
Crucially, ProtAI is not a “tools” company. Like Dafna Koller’s Insitro, they chose to own the drug pipeline, not just sell the platform. Their lead program targets estrogen receptors–positive breast cancer. For a completely novel mechanism with little prior data, they reached a near–development candidate in under 12 months—vs ~3 years by industry standards—and plan to enter the clinic in 2026 in the US and Israel.
Kiril’s biggest “anti–design pattern” in AI drug discovery:
Overhyping the AI itself. Pharma doesn’t buy models; they buy de‑risked drugs backed by hard data. Trust comes from experimental validation, fast iteration cycles, and clinical results—not from glossy benchmarks.
Listen in to learn how ProtAI thinks about moats, data, trust, and why speed of learning is the real superpower in AI-driven drug discovery.
Visit ProAI here
About me
I’m an entrepreneur, writer, host of Life Sciences Today podcast, ex-pharmatech founder, father of 4 and founder of OpenCRO.
OpenCRO specializes in threat modeling for FDA Cyber. 42 MedTech companies. 14 FDA clearances. 0 rework requests on threat model submissions.
You have threats to your patients, your clinical data, your reputation, your commercial operation, and your engineering roadmap
You shouldn’t be discovering your cybersecurity gaps during FDA review.
We find them first, document them the way FDA reviewers expect — and deliver to you a cost-effective, prioritized countermeasure plan.
Fixed price. Fixed outcome.
Book a call to see if you’re a good fit for OpenCRO.