Alex Cherucheril

Every drug you’ve heard of for arthritis, Crohn’s, psoriasis, or colitis works by blocking one protein: TNF-α. Humira, the best-selling drug in pharmaceutical history, is an antibody built to grab onto it. That single target generates $43 billion a year in drug sales. AI can now design entirely new proteins from scratch. On some targets, the results are startling. 83 out of 94 AI-designed proteins bound their target in real lab experiments. Not computer predictions. Actual molecules, actual test tubes. Then the same AI tried TNF-α. Zero out of 54 designs worked. This review catalogs real wet lab results across 200+ protein design tasks. The honest answer is beautiful and frustrating. AI protein design is spectacular on some biology and completely blind to other biology, and until now, nobody had systematically mapped where those lines fall. The biology decides the outcome. Not the model.

I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!

While impressive in its own right, it’s worth remembering that predicting protein-drug interactions is like 0.01% of the drug discovery and development pipeline and the spare change part of it. Saying that that’s revolutionizing the process is like convincing someone, in Derek Lowe’s memorable words, that you’ve invented a revolutionary new car because its windows go up and down ten times faster.

Demis Hassabis: If you know the structure of a protein, the real question becomes—where will your drug bind, and what will it actually do? That’s where the next wave of AI comes in. Not just predicting structures, but modeling interactions, outcomes, and real biological impact. At Isomorphic Labs, this is already happening—with 17 active drug programs and partnerships with giants like Eli Lilly and Novartis. The goal? Scale that to 100. This is a fundamental shift in how medicine gets built. Instead of slow, expensive trial-and-error in wet labs, AI allows researchers to run thousands of hypotheses in silico—hundreds to thousands of times more efficiently. The wet lab becomes validation, not exploration. Drug discovery is turning into a computational problem. Faster cycles, smarter predictions, and potentially massive breakthroughs in human health.