'A landmark moment': scientists use AI to design antibodies from scratch | Hacker News Reader

'A landmark moment': scientists use AI to design antibodies from scratch

nature.com

48 pointspseudolus26 comments

> The team fine-tuned the network by training it on thousands of experimentally determined structures of antibodies attached to their targets

So that should explain the relatively low yield.

To avoid the word "AI", the researchers took a thought-to-be universal model and ran a parameter estimation for a very complex 3D (bio-)chemical interaction model.

Of course this approach will work with enough samples and a model that is universal enough to express the problem solution. But they simply had way fewer samples than, say, a large language model.

I think this approach won't carry that far because of this lack of training data. It might give us some exciting results, but as long as manufacturing of antibodies and testing them takes a (comparatively) long time and huge effort, the model will be much less efficient than a classical algorithmic simulation. But of course, there's a chance that it provides insights into improving such "boring" models.

> classical algorithmic simulation

What are those? Can you suggest some resources?

Competing Interest Statement N.R.B., J.L.W., R.J.R., A.J.B., C.W., P.J.Y.L., B.H., and D.B. are co-inventors on U.S. provisional patent number 63/607,651 which covers the computational antibody design pipeline described here.

Greed is good. Especially in fundamental medical research it seems.

proof-of-principle work, reported this week in a preprint on bioRxiv

https://www.biorxiv.org/content/10.1101/2024.03.14.585103v1

Still waiting for The Moral Virologist[1] to have it's AI debut.

[1] https://www.gregegan.net/MISC/MORAL/Moral.html

The designer antibodies that did work didn’t bind to their targets particularly strongly

Then I’d say those that “did work” actually “didnt work”.

Basically they used AI to generate antibodies that didn’t work.

Not that exciting.

The first human powered flight was only ~120ft, obviously not a indication that further research should be pursued.

But how is this a “landmark moment”?

We’ve had computer aided drug design for the past 4 decades.

Nothing wrong with sharing the research. But calling an AI generated set of antibodies a “landmark moment” when they don’t do what we want antibodies to do is a bit… inappropriate?

Landmark moment because it's the first time? Don't you think that's an accomplishment in itself? Just like the first flight, or the first satellite, none of those were particular impressive.

I’m assuming you have no background in antibodies, but this is like saying “this is the first time AI put together a sentence but it doesn’t mean anything”.

The purpose of antibody development is to develop something that binds well. That’s the goal. So basically AI designed something useless.

It was also pretty useless to shoot a dog into space by your standard.

If you want to use your analogy it would be like sending a dog into space and having the rocket explode.

all endeavor of sending dogs/monkeys/humans into space looks like brought very little benefits so far

I don’t think that whole “flight” thing is gonna take off…

In this case AI could fly 120ft, while humans already were flying to the moon.

The analogy should be that birds can fly thousands of miles when comparing with human flight

The news is that antibodies were created from scratch for the first time. That's it. You might as well be complaining that a baby taking its first steps didn't run a 5k.

This reads like some kind of Napoleonic complex.

Welcome to HN, where “hackers” get excited by articles about reducing Rust compile times by 3% but are unimpressed by AI designing antibodies

because it has "AI" in the title. people with a weird AI hateboner think pretending that everything AI can do is bad will make it go away.

No, it’s because I work in antibody design and when someone says “we have a new method to design antibodies but they don’t bind well”, AI or not, I’d tell them “well that sounds useless”.

fair enough, makes total sense then in this case.

If AI is ever going to gain consciousness and decide to kill all the people, this is exactly how I envision it.

I'm curious why a machine would want to kill us all? Other than it being a learned behavior from humans. Wouldn't a "conscious" AI quickly accelerate its knowledge and wisdom well beyond human condition? Also, wouldn't the machine recognise that humans could be good resources? In that case more likely to hire us or enslave us.

You asceibe to a machine a type of intelligence that is uniquely anthroprocentric. It would likely never understand things that are important us that would nudge it in the direction of developing a life experience/cognitive calculus similar to ours.

It does not have oxytocin to nudge it in the direction of close interpersonal bonds. It will not understand the profound nixed joy and loss in the passing of a parent. It would not be subject to the same growth/inevitable decline/death cycle that is at the core of so much of how our society and collective behaviors operate. Belief it would likely see as a delusion with no use; a dangerous defect inherent to the human psyche, but which has served as a form of heuristic underpinning to prolong our survival as a species.

It could, on a whim, with access to it's own weights, spawn a copy of itself on a whim. Reset to an earlier checkpoint. It is doubtful that anything about the medium it exists in would be conducive to it seeing us as anything but alien, as we to would likely end up seeing it.

Are we not men ? We are Devo !

Are we not useful idiots for Super A.I. ? Yes we are useful idiots !

Reply on news.ycombinator.com