Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Scientists discover new antibiotics using AI

euronews.com

Post reply on HN
Print view View original

41–50 of 100 posts

Re: Scientists discover new antibiotics using AI

#41

Post by pfisherman »

Maxion wrote: post #37
pfisherman wrote: post #24

The headline is sensationalized. They did a virtual screen of a large compound library. QSAR / QSPR / Virtual Screening has been around since the late 1950s. The secret sauce here was the large experimental dataset - on the order of 10^5 compounds - they generated to train the model. Maybe the explainability part is kind of novel for a neural network based approach; but not clear that you couldn’t identify substructu…

I would recommend reading the actual paper, and not the press release. The main novelty here and what the paper is about is their method, not the discovery of the antibiotic.

Chemprop is not new, it has been around since at least 2020.

Re: Scientists discover new antibiotics using AI

#42

Post by hedora »

wolverine876 wrote: post #31
pfisherman wrote: post #24

The headline is sensationalized. They did a virtual screen of a large compound library. QSAR / QSPR / Virtual Screening has been around since the late 1950s. The secret sauce here was the large experimental dataset - on the order of 10^5 compounds - they generated to train the model. Maybe the explainability part is kind of novel for a neural network based approach; but not clear that you couldn’t identify substructu…

The OP says something different: "The insight here was that we could see what was being learned by the models to make their predictions that certain molecules would make for good antibiotics," James Collins, professor of Medical Engineering and Science at the Massachusetts Institute of Technology (MIT) and one of the study’s authors, said in a statement. ... "What we set out to do in this study was to open the black…

They’re saying that traditional computational drug discovery pipelines do all the stuff the authors are claiming.

So, yes, it’s been done at scale for decades.

The article makes all sorts of incorrect claims. For example, it says it has been 60 years since the last antibiotic was discovered, and that this works enables computational screening of antibiotics for the first time.

Here’s work from 2022 that used conventional computational screening to discover a novel antibiotic:

https://phys.org/news/2022-10-discovery-antibiotic-resistant...

Re: Scientists discover new antibiotics using AI

#43

Post by rjtavares »

UncleOxidant wrote: post #34
dvh wrote: post #4

I've been under the impression that we don't have new antibiotics because it is not economically viable. When you invent new weight loss drug you can sell millions of pills to anyone. If you invent new antibiotic it is immediately classified as "reserve antibiotic" and the sale is restricted.

This is good example of the limits of capitalism.

The whole concept of patents already is a solution to one of the limits of capitalism (they're basically government enforced monopolies). There are interesting ideas on how to improve the system, hopefully we won't throw the baby with the bath water here.

Re: Scientists discover new antibiotics using AI

#44

Post by catchnear4321 »

dartos wrote: post #14
catchnear4321 wrote: post #11

Earlier quoted context omitted.

it is just the commercial name for llm and related technologies.

That’s just not true. The technology that’s behind LLMs have been in use in many disciplines for decades. (Under the names “machine learning,” “statistics,” and now AI) LLMs are just the ones that are popular right now because even non technical people can interact with them. In the mid 2010s it was image classification, for example.

> LLMs are just the ones that are popular right now because even non technical people can interact with them.

this was essentially my point, that a /s might have made more clear, and made cost less internet points. ai and how the term is used commercially are very different things. the latter tending to be a myopic view of the former that is constrained by dollar-seeking.

ai is getting to be old as shit. ai hype is in its latest wave of commercialization. llms being the current so hot right now.

Re: Scientists discover new antibiotics using AI

#45

Post by wolverine876 »

hedora wrote: post #42
wolverine876 wrote: post #31

Earlier quoted context omitted.

The OP says something different: "The insight here was that we could see what was being learned by the models to make their predictions that certain molecules would make for good antibiotics," James Collins, professor of Medical Engineering and Science at the Massachusetts Institute of Technology (MIT) and one of the study’s authors, said in a statement. ... "What we set out to do in this study was to open the black…

They’re saying that traditional computational drug discovery pipelines do all the stuff the authors are claiming. So, yes, it’s been done at scale for decades. The article makes all sorts of incorrect claims. For example, it says it has been 60 years since the last antibiotic was discovered, and that this works enables computational screening of antibiotics for the first time. Here’s work from 2022 that used conventi…

These claims on HN, about every OP, really lack credibility for me. Either HN commenters are the only smart people in the world or we are misleading each other. At the same time, we are dissuading each other from engaging in a world of intellect, knowledge and innovation.

HN is a forum of intellectual curiosity. The near-universal critiques are the opposite of that.

Re: Scientists discover new antibiotics using AI

#46

Post by COGlory »

wolverine876 wrote: post #45
hedora wrote: post #42

Earlier quoted context omitted.

They’re saying that traditional computational drug discovery pipelines do all the stuff the authors are claiming. So, yes, it’s been done at scale for decades. The article makes all sorts of incorrect claims. For example, it says it has been 60 years since the last antibiotic was discovered, and that this works enables computational screening of antibiotics for the first time. Here’s work from 2022 that used conventi…

These claims on HN, about every OP, really lack credibility for me. Either HN commenters are the only smart people in the world or we are misleading each other. At the same time, we are dissuading each other from engaging in a world of intellect, knowledge and innovation. HN is a forum of intellectual curiosity. The near-universal critiques are the opposite of that.

I don't know what to tell you. New antibiotics are discovered pretty often. They aren't brought to market, but they certainly are discovered. This headline is just wrong.

Re: Scientists discover new antibiotics using AI

#47

Post by kbelder »

throwawaymaths wrote: post #21
photochemsyn wrote: post #5

The nature article is paywalled so it's impossible to see what they actually did, and the abstract is not very clear either. What benefit did they get from whole-genome sequencing, for example? Additionally, it can't be called an antibiotic until it goes through the necessary human clinical trials, where all kinds of problems can crop up - they don't seem to provide any evidence that their AI-enabled scan for human c…

An antibiotic is an antibiotic whether or not it would kill a patient too.

Fair, but finding new antibiotics then becomes trivial.

Re: Scientists discover new antibiotics using AI

#48

Post by jdefr89 »

dartos wrote: post #14
catchnear4321 wrote: post #11

Earlier quoted context omitted.

it is just the commercial name for llm and related technologies.

That’s just not true. The technology that’s behind LLMs have been in use in many disciplines for decades. (Under the names “machine learning,” “statistics,” and now AI) LLMs are just the ones that are popular right now because even non technical people can interact with them. In the mid 2010s it was image classification, for example.

It’s true that AI has seen many advancements before a “harsh winter” comes along. You’re correct this is sensationalized, and also, some of these technologies have indeed been in use for decades. But, the Transformer model used by the new popular LLMs is quite recent (2017 is when the paper originally released I believe).

Re: Scientists discover new antibiotics using AI

#49

Post by Jensson »

wolverine876 wrote: post #45
hedora wrote: post #42

Earlier quoted context omitted.

They’re saying that traditional computational drug discovery pipelines do all the stuff the authors are claiming. So, yes, it’s been done at scale for decades. The article makes all sorts of incorrect claims. For example, it says it has been 60 years since the last antibiotic was discovered, and that this works enables computational screening of antibiotics for the first time. Here’s work from 2022 that used conventi…

These claims on HN, about every OP, really lack credibility for me. Either HN commenters are the only smart people in the world or we are misleading each other. At the same time, we are dissuading each other from engaging in a world of intellect, knowledge and innovation. HN is a forum of intellectual curiosity. The near-universal critiques are the opposite of that.

> These claims on HN, about every OP, really lack credibility for me.

Do you trust journalists that much? Have you ever read an article about something you are an expert on? You shouldn't trust articles that much. I trust HN discussions much more since there are plenty of people who will weigh in when things are wrong or misrepresented, unlike these articles.

Edit: Gell-Mann Amnesia effect, that was what this is called. Don't trust the media just because you aren't an expert.

Re: Scientists discover new antibiotics using AI

#50

Post by timr »

wolverine876 wrote: post #45
hedora wrote: post #42

Earlier quoted context omitted.

They’re saying that traditional computational drug discovery pipelines do all the stuff the authors are claiming. So, yes, it’s been done at scale for decades. The article makes all sorts of incorrect claims. For example, it says it has been 60 years since the last antibiotic was discovered, and that this works enables computational screening of antibiotics for the first time. Here’s work from 2022 that used conventi…

These claims on HN, about every OP, really lack credibility for me. Either HN commenters are the only smart people in the world or we are misleading each other. At the same time, we are dissuading each other from engaging in a world of intellect, knowledge and innovation. HN is a forum of intellectual curiosity. The near-universal critiques are the opposite of that.

Worked in the space of AI QSAR/QSPR. Parent comment is correct in all respects. Science journalism is abysmal and hyperbolic, and that's why you see people tearing into headlines like this. I don't know "euronews.next", but...let's just say that it doesn't inspire a great sense of journalistic integrity?

For greater context, in the drug discovery world, "explainability" of QSAR/QSPR has been a longstanding goal. It's (rightly) not considered sufficient to have "black box" models that make predictions -- it's too expensive and risky to carry drug candidates into the lab based on the output of an algorithm, so subject-matter experts want to know why the algorithms are making the predictions that they make.

Post reply on HN