“Even with things like monoclonal antibodies, which are very specific, there are people who went into critical care in phase one clinical trials because of some unexpected side effect, off-target binding. So you really have to do all the trials before you can release it on the market. AI could be very fast at its computational work, but the experimental work is still going to be the rate-limiting factor.”
The Peterman Post · Software Engineering & Infra · November 2026
Context
Ryan Peterman asked what would happen if AI were a perfect oracle that could hand you a molecule that binds the target. Ramakrishnan, a Nobel-winning structural biologist, argues the physical-world steps after the computation still set the pace. It's a direct counterweight to claims that AI will rapidly compress drug development.
Transcript
Yes, that's the whole premise, is that it can speed up and make that initial filter much more efficient. So that's the hope. And I think, you know, we have to wait a few years and see how it's panning out.
But even if that was a perfect, let's say it was an oracle. It could just give you something that you know binds. There's still all these downstream steps in the physical. Oh, yeah,
definitely. You know, because I can tell you, even with things like monoclonal antibodies, which are very specific, there are people who went into critical care in phase one clinical trials because of some unexpected side effect, you know, off-target binding. So you really have to do all the trials before you can release it on the market. So it's not, you know, AI could be very fast at its computational work, but the experimental work is still going to be the rate-limiting factor. Human biology is complicated. So, you know, it's not, I don't know if it's possible to predict all of those, you know, potential side effects and the kinetics of the drug, how quickly is the drug excreted. You know, these are all things you have to do experimentally at the moment. I mean, in some distant future, you know, maybe we'll reach a world where these things can be predicted, but I don't think we're there today.
Is there enough slots for all the people who have drugs that want to do clinical testing? Well,
they have to fund it and find the patients and so on. So that's the limiting factor. It's not, I mean, the FDA, if you propose, you know, if you have sufficient animal data and, you know, some, you know, animal safety data and a clear rationale, the FDA will say, okay, you can perform it, you can…
In other fields like math and computer science that are more closed, we see that these models are drastically changing the frontier and the state of the art. Do you expect a similar down the road in molecular biology?
More from The Peterman Post
