Hacker News Reader

Comment by ProjectArcturis | Hacker News Reader

It won't? AI-powered drug screening has definitely been overhyped, but in the longer term highly accurate protein structure modelling should let us understand protein-protein interactions and provide new opportunities for intervention.

View on HN

That has been a long-stated claim in the field that has repeatedly not shown to be useful in any sort of "engineering" or "medical treatment" sense.

Hrmm, I think I was expecting a different type of disagreement when you said he was wrong on facts.

I'm sure you know your stuff, and that you have a lot of experience with proteins that haven't helped with drug discovery or engineering, but it sounds like this is indeed a mismatch between predictions rather than facts.

It very well could be the case that speeding up certain problems by multiple orders of magnitude really does help with drug discovery, and this isn't factually inconsistent with the fact that solving those problems hasn't turned out to be useful so far in this area.

If you can discover a drug faster and that drug is as useful as dirt does it matter?

This isn't my field but I could grab a bunch of random jars off a shelf and pour them into capsules. No matter how fast I can do this won't improve medical outcomes for patients.

you just described how modern high content screening, which has been one of the most useful techniques for finding drug leads, works. Since lead-finding is a bottleneck in the drug discovery process, it has been highly effective because it can measure things that are not currently computationally accessible.

is a bottleneck != is THE bottleneck

It's the first bottleneck. Generally, if you can't pass the first bottleneck, all the remaining bottlenecks don't matter.

I don't follow.

Wasn't your claim that the AI process hasn't been shown to generate a useful drug [1]?

[1]: https://news.ycombinator.com/item?id=33720915

Don’t forget that we already have highly accurate protein structure modelling. AlphaFold adds to that but it’s not like it’s something radically new. Proteins involved in diseases we care about have been extensively studied.

Except a blackbox that simply spits out a resultant folded protein doesn't actually improve our understanding of anything. Are we just going to fuzz the blackbox with different protein combos hoping to find something useful? In that case, aren't we just more likely to find some stupid error in the ML predictor?

Reply on news.ycombinator.com