10 ms·

Scientists use ML to find an antibiotic able to kill superbugs in mice

438 points by adventured 7y ago

entee 7y ago
This paper shows what can be done when you carefully run an ML program alongside a wet lab experimental program tailored to feed back into the ML program. The results end up far more interesting than some recent "ML aided drug discovery" papers in that they actually discovered a drug that functions very differently than known antibiotics (AB). Even though the structures that came out look AB-like, they work differently than known ABs, probably by disrupting the pH gradient across the cell membrane. Other ABs might do the same as part of their activity, but work better under different conditions than this one. The result is an innovative structure, and a molecule that can hit resistant strains. Combining ML and wet lab is the real way we'll get new drugs. You need to regularly check in with a high content ground truth or you'll come up with either uninteresting or useless results. I'm a bit biased though, that's what we do at my company ;)
- chrisgd 7y ago
  Can you help me understand what is the difference between what you are doing and what people like Certara and Simulations Plus are doing? The simulation software market is really fascinating
  - entee 7y ago
    Sure! To a first approximation they both provide software packages to pharma who then apply those to their efforts. In our case we are building an integrated system with a biochemical component and a computational component. We will then seek out molecules to difficult targets ourselves because we believe the integration gives us a competitive advantage.
- spitfire 7y ago
  Not just in drug discovery but in most interesting industries. Using ML as either a human/cyborg aid or ML+real world ground truth is a secret superpower that I'm surprised more people don't know about. I'm glad they don't.
  - timClicks 7y ago
    They do. It's just easier to add extra CPU cores to a ML pipeline than another person.
  - IAmEveryone 7y ago
    As a less cynical observation than a sister comment, I think this is partly due to how current tools work, or possibly even inherent to the technology. Just look at the number of ML examples where you never even get to see a single example of the data you are working with. Interactivity is even rarer, certainly because it doesn’t scale, but maybe also for the somewhat pedestrian reason that interactive UIs doesn’t come natural for either the tools and/or the people using them. Then, there’s this issue of “explainability”: if you want to direct some generator you need to find out how the concepts you want to work with are encoded in intermediary layers. To be fair, all this isn’t much of a secret, and there are quite a few projects doing interesting things. Magenta comes to mind, or GANBreeder.
  - sgt101 7y ago
    One challenge here is seen in the types of articles researhers in comp sci have published over the years. I believe that there has been a massive decline in case studies and field trials of software and a rise in analytic and empirical algorithm development. To unlock the super power for people we will need more good hci and the wheel will have to turn in research practice once more.
- xzel 7y ago
  People have been doing this exact thing for two decades at least but obviously with less computing power. There's literally nothing new about the idea. The real trick is being incredibly lucky and finding something that actually works in humans after multiple trials. I'm sure you know this based on your comment and this isn't really directed at you (truly wish you best of luck, I really hope the computing power and skill we have saves lives) but I absolutely hate these types of articles, which I've seen becoming more frequent the past year. AI + scientific challenge + possibility = puff article. This is the type of semi-hysterical reporting I expect from a local news station.
  - entee 7y ago
    I share some of your skepticism, but I think this particular paper is worth a little less cynicism. It’s true that statistical analysis has always helped inform medicinal chemistry efforts. They used to be called QSAR ;) What’s different here is the scale of information that was processed. As you point out the computational power and therefore the algorithms we can apply has enabled different experiments which I very much hope will yield better and more abundant medicines. Papers like this are a sign of these methods bearing fruit!
    - xzel 7y ago
      I totally agree, and I think my comment could be summarized as such from another angle. I think the recent Coronavirus + AI = cure spam has me a little jaded. I'll add this, this paper is really well done and like I said I really hope this works in humans!
  - headmelted 7y ago
    I’m usually also skeptical of getting excited about successful animal trials for new medications, but with this I’m actually leaning more on the side of optimism. With cancer treatments and antivirals in mice, we’re not so much targeting the pathogen as targeting the host immune system in the hopes it ends up nerfing the intended target (tumor/virus/whatever). Given that the compound seems effective against C. Difficile (even if it’s in mice), I’d expect it to work elsewhere. Of course, I’m not a doctor and have no idea what I’m talking about so grain of salt required.
- naresh_xai 7y ago
  Exactly why we need causal reasoning/causal proof alongwith ML
  - entee 7y ago
    In this case I’m not sure if the reasoning for what molecules work would make sense even if we had an “oracle” to explain it. Why a molecule works is a complicated interplay of fundamental physics and emergent properties. The explanation is likely not human interpretable without a ton of equations in the first place. That said in broader medical cases of ML where particular symptoms and measurements drive a diagnosis for example, I tend to agree.
    - IAmEveryone 7y ago
      I’m not sure about the exact mechanisms of drug efficiency. But there are well-known patterns in molecular biology, auch as specific sequences in proteins reliably forming specific 3D structures. If these models produce novel but (somewhat) effective structures, it must be because they pick up on less obvious patterns in the data. To be able to describe these would seem to super effective.
koolba 7y ago
> That is an especially pressing challenge in the development of new antibiotics, because a lack of economic incentives has caused pharmaceutical companies to pull back from the search for badly needed treatments. Each year in the U.S., drug-resistant bacteria and fungi cause more than 2.8 million infections and 35,000 deaths, with more than a third of fatalities attributable to C. diff, according to the the Centers for Disease Control and Prevention. How big does the market have to be to commercially viable for research and development? Nearly 3M potential patients at a couple hundred dollars per course is nearing a $1B/year.
- westurner 7y ago
  The second-order costs avoided by treatments developed so innovatingly could be included in a "value to society" estimation. "Acknowledgements" lists the grant funders for this federally-funded open access study. "A Deep Learning Approach to Antibiotic Discovery" (2020) https://doi.org/10.1016/j.cell.2020.01.021 https://doi.org/10.1016/j.cell.2020.01.021 > Mutant generation > Chemprop code is available at: https://github.com/swansonk14/chemprop https://github.com/swansonk14/chemprop > Message Passing Neural Networks for Molecule Property Prediction > A web-based version of the antibiotic prediction model described herein is available at: http://chemprop.csail.mit.edu/ http://chemprop.csail.mit.edu/ > This website can be used to predict molecular properties using a Message Passing Neural Network (MPNN). In order to make predictions, an MPNN first needs to be trained on a dataset containing molecules along with known property values for each molecule. Once the MPNN is trained, it can be used to predict those same properties on any new molecules.
- kirrent 7y ago
  At least one disincentive is that if you do find an amazing new antibiotic effective against certain strains of antibiotic resistant bacteria, antibiotic stewardship means the medical community will try and use it only where necessary to slow any adaptation to the new drug. That makes your potential patient population much smaller.
  - rkangel 7y ago
    If we develop new ways to invent anti-biotics, such that we have a reasonable expectation of new ones being produced yearly, then maybe anti-biotic stewardship will become less important.
ErotemeObelus 7y ago
This method might be generalizable to prions.
antipaul 7y ago
Thank goodness it was "aided by machine learning". Otherwise, would anybody be reading this article ;)?
- wyattpeak 7y ago
  We're techies, not doctors. It's the computational solutions that interest us, only secondarily the underlying problems being solved. Thankfully, it doesn't matter a whit whether or not a bunch of programmers read about medical advancements on their lunch break.
  - kharak 7y ago
    You should realised that not everyone here identifies as techi. I for one care about interesting problems and ideas and ignore most of the pure tech posts. From other posts I've seen, there are people from quit different occupations here.
  - bjonnh 7y ago
    There are a few scientists working in drug discovery here (I'm one of them), and quite a few from other domains from what I've read. We have philosophers that intervene regularly as well. Not just techies!
DoreenMichele 7y ago
...in mice. It would be nice if we spent as much time, money and attention on figuring out prevention. Inadequate hygiene infrastructure (like toilets) in developing areas is part of the problem here. But addressing that isn't as exciting to people as finding a cure for a super bug. If we really want to fix this, that needs to change.
- bravoetch 7y ago
  We do.
  - DoreenMichele 7y ago
    I can't think of the last time I saw a headline on the front page of HN about the importance of solving open defecation. I see headlines about creating new treatments for super bugs pretty regularly. Open defecation costs $US260b globally https://www.news.com.au/world/breaking-news/open-defecation-costs-us260b-globally/news-story/d3cdd105a8d2c3dc6002f626e9d791b7 https://www.news.com.au/world/breaking-news/open-defecation-... Price to pay: Antibiotic-resistant infections cost $2 billion a year http://www.cidrap.umn.edu/news-perspective/2018/03/price-pay-antibiotic-resistant-infections-cost-2-billion-year http://www.cidrap.umn.edu/news-perspective/2018/03/price-pay... We basically know how to solve open defecation: build toilets and sewage systems. I think we could largely solve open defecation in five years if we really wanted to. But we aren't throwing resources at it like crazy because it isn't a "sexy" issue and because first world wealthy people don't see it as directly relevant to their lives. They see it as "being nice to poor people in developing nations," not as an urgent global priority because it's fueling the creation of antibiotic resistant infections which can be exported within 24 hours to their country by someone jumping on a plane before they are symptomatic. And I really can't recall ever seeing a headline explicitly linking the two things. But they are linked. http://resistancecontrol.info/2017/prevention-first-tackling-amr-through-water-sanitation-and-hygiene/ http://resistancecontrol.info/2017/prevention-first-tackling...
    - pkaye 7y ago
      Bill Gates did a lot of funding for developing low cost toilets for developing countries. There is a documentary called "Bill Gates Mind" on Netflix that talks about it.
tjchear 7y ago
I'm not an expert in ML/Biology, but I wonder if doing so won't completely eliminate all of humankind's diseases, but shift the battle from one between humans and bacteria/viruses, to one where ML takes the place of humans by proxy (say we let this ML vs superbugs play out over centuries). I wonder to what direction evolutionary pressure in the face of ML would take bacteria. Perhaps a super smart bug.
- ganarajpr 7y ago
  Thats a very interesting thought. I am not sure evolution can compete with artificial evolution ( ML ). But we never know :). Fingers crossed that ML can always beat evolution.
  - michaelscott 7y ago
    That's also a very interesting thought. The general narrative with general AI is that natural evolution in humans is too slow, we won't be able to catch up mentally with such an AI and this is a bad thing. But the flip side is that natural evolution that is damaging to us (bacteria and viruses) also won't be able to evolve fast enough.
- Rochus 7y ago
  Molecular biology is not that easy. Selecting a promising molecule is only a small fraction of the work. If you know the molecule you still have to be able to replicate and bring it to the place where it is required. So even if AI can support the process still a lot of plain old biochemistry is needed. AI is just one of many tools. And as you can read in the article a lot of trials are required by the regulatory authorities before the drug can be applied to humans.
  - antirez 7y ago
    True but it's interesting that once you have an oracle that given a molecule provides you with an antibiotic score, you can show it directly molecules that are easy to produce and or likely safe for humans. This should help significantly.
throwGuardian 7y ago
What's becoming apparent is the fundamental nature of applied mathematics and ML/AI/DL to our future. Our education system needs to adapt, include this as a mandatory part of a college education (BRIC countries are including this at the high school level). A degree in pure AI/ML without mastery over impactful problems and it's underlying science, is not the ideal future. Every chemist, biologist, ... should be proficient in ML/DL
- mantap 7y ago
  The field of ML is moving too fast to be taught in schools. It's better to teach its foundations (linear algebra, etc) which are well developed and static.
thedance 7y ago
It's interesting here that the breakthrough seems to come from the computer search not suffering from epistemological block. It doesn't think, therefore it is also not bound by conventional patterns of thought, such as the bias a researcher may show when thinking that this known compound is not an antibiotic, when it is in fact an antibiotic compound.
logifail 7y ago
Decades ago when I worked in a lab, "high-throughput screening" was definitely on the list of buzzwords. Given what I saw back then, I'm struggling to understand how there could possibly be "a library [..] of 6,111 molecules at various stages of investigation for human diseases" (a.k.a. "Drug Repurposing Hub") which hasn't already been partially or fully screened for interesting antibiotic activity. Could it be there's more fame (and funding) in a project where you can publish a paper and get headlines about "ML" and "superbugs", than in actually testing a library of existing compounds to see if any of them kill E. coli?
- anonsivalley652 7y ago
  For DD, it seems initial screening of as many phages, microbes and compounds as possible using highly-automated brute force might be plenty efficient to test their effectivenesses against every horrible, resistant and opportunistic pathogen for candidate identification. Maybe flying drones out to collect samples in as many random places (public places, restrooms, dirt and even more random places) as possible, generating more samples than a team of humans ever could. There doesn't seem to be any "One True Way," but a holistic synthesis of collection, identification and selection methods.
  - logifail 7y ago
    > highly-automated brute force might be plenty efficient Hasn't this been going on in one form or another for many decades? When I was in this field (20+ years ago) I got to visit labs at Glaxo Wellcome, SmithKline Beecham, Zeneca and so on. Even back then they were proudly showing off lab robots which allowed them to run large-scale screening experiments. Not sure any of this stuff is quite as revolutionary as it looks.
    - kokey 7y ago
      I did a stint at GSK about 13 years ago. The large scale mechanised experiments also produced big databases of compounds and their properties and mechanisms to make these accessible to researchers and build bigger clusters of systems to run models on. There was a lot of talk of ML but implementation was nowhere near the scale we see nowadays. I think what has changed significantly is the scale, second to that the methods including algorithms and methods to feed back real data into it. It's an evolutionary improvement but I sense it's one of those things where one lucky step in the evolution could make a big difference in productivity.
tasubotadas 7y ago
Here is the code repository for the model https://github.com/chemprop/chemprop https://github.com/chemprop/chemprop . I am surprised by the (high) quality of the code published there. It is quite a rare case in the academic world :).
- phobar 7y ago
  They are funded by several big pharma companies (from which they have learned a lot in terms of QSAR). One main point of the collaboration is the creation of production ready toolkits.
loopasam 7y ago
A medicinal chemist take on it: https://blogs.sciencemag.org/pipeline/archives/2020/02/20/machine-learning-for-antibiotics https://blogs.sciencemag.org/pipeline/archives/2020/02/20/ma...
lngnmn1 7y ago
Why, yes, pattern recognition is what ML really is and should be used for. Aside from pattern recognition everything else is a new astrology, with estimated (straight from one's ass) probabilities instead of planets and constellations.
allovernow 7y ago
A lot of the more cynical commenters here are misunderstanding the breakthroughs that have lead to the novel discovery, and underpin the emerging ML revolution that we are just beginning to witness. Yes, partly the results of this study are due to increased compute, but I'd say that's only about 50% of the secret sauce. The other 50% is attributable to many very recent developments in the field ML which are gradually coming together in solutions to a range of problems - deep learning, convolutional networks, new architectures (GAN, transformer, RNN, LSTM, autoencoders, etc), regularization techniques, gradient control, hyperparameter optimization...the list is quite long, and every SOTA neural network inevitably incorporates a large proportion of these small steps towards the massive leaps that are being taken in the applied ML space. The concepts of perceptrons and gradient descent may be 50+ years old, but an entire body of theory has been explosively developed in the last decade, such that comparing modern ML to what was being done just a decade ago is akin to the difference between, say, programming theory now and 100 years ago. And for the same reason - just as compute advances enabled development of computer science, so to have recent hardware advances unlocked a new level of machine learning. But I'd like to point out that, even moreso than software development, very little of the grand breakthroughs we will soon see will be possible without multidisciplinary domain knowledge. It is very difficult to effectively apply ML without a solid technical understanding of the properties of the applied data space, which for real world applications are constrained by physical laws and represented and communicated best by mathematical descriptions. ML engineering is a generalist's game - and what we are going to find is that the most successful ML engineers come from broadly applicable, math heavy backgrounds - physics in particular, electrical engineering, to a lesser degree mathematics, etc - because ultimately training a neural network comes down to adequately sampling a problem space and curating data with an intuition which is most ideally developed by the study of mathematics. It is a very general view of the world which is difficult to communicate to someone who is not experienced with higher math. The current wave of applied ML startups will see a high rate of failure - because ML is still being treated as an extension of programming, in the sense that you expect to be able to hire a bunch of pure developers to translate a specialist's knowledge into code. But this emerging field is different, the few startups that succeed in the applied ML space will be those that are able to find the rare domain experts who have picked up ML along with their math and science experience. There will effectively emerge two classes of ML engineers with substantially different levels of compensation - the coveted generalists who have cross-pollinated with math heavy disciplines, and the rest.
dom96 7y ago
I saw news about this on The Guardian and it was the first time in a while where I was truly amazed by an ML-related discovery. This kind of thing is incredibly exciting and I hope we continue to see similar discoveries in ever more wider fields.
DeonPenny 7y ago
Knew this happen eventually. ML drugs will be huge

438 points by adventured 7y ago

This paper shows what can be done when you carefully run an ML program alongside a wet lab experimental program tailored to feed back into the ML program. The results end up far more interesting than some recent "ML aided drug discovery" papers in that they actually discovered a drug that functions very differently than known antibiotics (AB). Even though the structures that came out look AB-like, they work differently than known ABs, probably by disrupting the pH gradient across the cell membrane. Other ABs might do the same as part of their activity, but work better under different conditions than this one. The result is an innovative structure, and a molecule that can hit resistant strains. Combining ML and wet lab is the real way we'll get new drugs. You need to regularly check in with a high content ground truth or you'll come up with either uninteresting or useless results. I'm a bit biased though, that's what we do at my company ;)

chrisgd 7y ago
Can you help me understand what is the difference between what you are doing and what people like Certara and Simulations Plus are doing? The simulation software market is really fascinating
- entee 7y ago
  Sure! To a first approximation they both provide software packages to pharma who then apply those to their efforts. In our case we are building an integrated system with a biochemical component and a computational component. We will then seek out molecules to difficult targets ourselves because we believe the integration gives us a competitive advantage.
spitfire 7y ago
Not just in drug discovery but in most interesting industries. Using ML as either a human/cyborg aid or ML+real world ground truth is a secret superpower that I'm surprised more people don't know about. I'm glad they don't.
- timClicks 7y ago
  They do. It's just easier to add extra CPU cores to a ML pipeline than another person.
- IAmEveryone 7y ago
  As a less cynical observation than a sister comment, I think this is partly due to how current tools work, or possibly even inherent to the technology. Just look at the number of ML examples where you never even get to see a single example of the data you are working with. Interactivity is even rarer, certainly because it doesn’t scale, but maybe also for the somewhat pedestrian reason that interactive UIs doesn’t come natural for either the tools and/or the people using them. Then, there’s this issue of “explainability”: if you want to direct some generator you need to find out how the concepts you want to work with are encoded in intermediary layers. To be fair, all this isn’t much of a secret, and there are quite a few projects doing interesting things. Magenta comes to mind, or GANBreeder.
- sgt101 7y ago
  One challenge here is seen in the types of articles researhers in comp sci have published over the years. I believe that there has been a massive decline in case studies and field trials of software and a rise in analytic and empirical algorithm development. To unlock the super power for people we will need more good hci and the wheel will have to turn in research practice once more.
xzel 7y ago
People have been doing this exact thing for two decades at least but obviously with less computing power. There's literally nothing new about the idea. The real trick is being incredibly lucky and finding something that actually works in humans after multiple trials. I'm sure you know this based on your comment and this isn't really directed at you (truly wish you best of luck, I really hope the computing power and skill we have saves lives) but I absolutely hate these types of articles, which I've seen becoming more frequent the past year. AI + scientific challenge + possibility = puff article. This is the type of semi-hysterical reporting I expect from a local news station.
- entee 7y ago
  I share some of your skepticism, but I think this particular paper is worth a little less cynicism. It’s true that statistical analysis has always helped inform medicinal chemistry efforts. They used to be called QSAR ;) What’s different here is the scale of information that was processed. As you point out the computational power and therefore the algorithms we can apply has enabled different experiments which I very much hope will yield better and more abundant medicines. Papers like this are a sign of these methods bearing fruit!
  - xzel 7y ago
    I totally agree, and I think my comment could be summarized as such from another angle. I think the recent Coronavirus + AI = cure spam has me a little jaded. I'll add this, this paper is really well done and like I said I really hope this works in humans!
- headmelted 7y ago
  I’m usually also skeptical of getting excited about successful animal trials for new medications, but with this I’m actually leaning more on the side of optimism. With cancer treatments and antivirals in mice, we’re not so much targeting the pathogen as targeting the host immune system in the hopes it ends up nerfing the intended target (tumor/virus/whatever). Given that the compound seems effective against C. Difficile (even if it’s in mice), I’d expect it to work elsewhere. Of course, I’m not a doctor and have no idea what I’m talking about so grain of salt required.
naresh_xai 7y ago
Exactly why we need causal reasoning/causal proof alongwith ML
- entee 7y ago
  In this case I’m not sure if the reasoning for what molecules work would make sense even if we had an “oracle” to explain it. Why a molecule works is a complicated interplay of fundamental physics and emergent properties. The explanation is likely not human interpretable without a ton of equations in the first place. That said in broader medical cases of ML where particular symptoms and measurements drive a diagnosis for example, I tend to agree.
  - IAmEveryone 7y ago
    I’m not sure about the exact mechanisms of drug efficiency. But there are well-known patterns in molecular biology, auch as specific sequences in proteins reliably forming specific 3D structures. If these models produce novel but (somewhat) effective structures, it must be because they pick up on less obvious patterns in the data. To be able to describe these would seem to super effective.

koolba 7y ago

> That is an especially pressing challenge in the development of new antibiotics, because a lack of economic incentives has caused pharmaceutical companies to pull back from the search for badly needed treatments. Each year in the U.S., drug-resistant bacteria and fungi cause more than 2.8 million infections and 35,000 deaths, with more than a third of fatalities attributable to C. diff, according to the the Centers for Disease Control and Prevention. How big does the market have to be to commercially viable for research and development? Nearly 3M potential patients at a couple hundred dollars per course is nearing a $1B/year.

westurner 7y ago
The second-order costs avoided by treatments developed so innovatingly could be included in a "value to society" estimation. "Acknowledgements" lists the grant funders for this federally-funded open access study. "A Deep Learning Approach to Antibiotic Discovery" (2020) https://doi.org/10.1016/j.cell.2020.01.021 https://doi.org/10.1016/j.cell.2020.01.021 > Mutant generation > Chemprop code is available at: https://github.com/swansonk14/chemprop https://github.com/swansonk14/chemprop > Message Passing Neural Networks for Molecule Property Prediction > A web-based version of the antibiotic prediction model described herein is available at: http://chemprop.csail.mit.edu/ http://chemprop.csail.mit.edu/ > This website can be used to predict molecular properties using a Message Passing Neural Network (MPNN). In order to make predictions, an MPNN first needs to be trained on a dataset containing molecules along with known property values for each molecule. Once the MPNN is trained, it can be used to predict those same properties on any new molecules.
kirrent 7y ago
At least one disincentive is that if you do find an amazing new antibiotic effective against certain strains of antibiotic resistant bacteria, antibiotic stewardship means the medical community will try and use it only where necessary to slow any adaptation to the new drug. That makes your potential patient population much smaller.
- rkangel 7y ago
  If we develop new ways to invent anti-biotics, such that we have a reasonable expectation of new ones being produced yearly, then maybe anti-biotic stewardship will become less important.

ErotemeObelus 7y ago

This method might be generalizable to prions.

antipaul 7y ago

Thank goodness it was "aided by machine learning". Otherwise, would anybody be reading this article ;)?

wyattpeak 7y ago
We're techies, not doctors. It's the computational solutions that interest us, only secondarily the underlying problems being solved. Thankfully, it doesn't matter a whit whether or not a bunch of programmers read about medical advancements on their lunch break.
- kharak 7y ago
  You should realised that not everyone here identifies as techi. I for one care about interesting problems and ideas and ignore most of the pure tech posts. From other posts I've seen, there are people from quit different occupations here.
- bjonnh 7y ago
  There are a few scientists working in drug discovery here (I'm one of them), and quite a few from other domains from what I've read. We have philosophers that intervene regularly as well. Not just techies!

DoreenMichele 7y ago

...in mice. It would be nice if we spent as much time, money and attention on figuring out prevention. Inadequate hygiene infrastructure (like toilets) in developing areas is part of the problem here. But addressing that isn't as exciting to people as finding a cure for a super bug. If we really want to fix this, that needs to change.

bravoetch 7y ago
We do.
- DoreenMichele 7y ago
  I can't think of the last time I saw a headline on the front page of HN about the importance of solving open defecation. I see headlines about creating new treatments for super bugs pretty regularly. Open defecation costs $US260b globally https://www.news.com.au/world/breaking-news/open-defecation-costs-us260b-globally/news-story/d3cdd105a8d2c3dc6002f626e9d791b7 https://www.news.com.au/world/breaking-news/open-defecation-... Price to pay: Antibiotic-resistant infections cost $2 billion a year http://www.cidrap.umn.edu/news-perspective/2018/03/price-pay-antibiotic-resistant-infections-cost-2-billion-year http://www.cidrap.umn.edu/news-perspective/2018/03/price-pay... We basically know how to solve open defecation: build toilets and sewage systems. I think we could largely solve open defecation in five years if we really wanted to. But we aren't throwing resources at it like crazy because it isn't a "sexy" issue and because first world wealthy people don't see it as directly relevant to their lives. They see it as "being nice to poor people in developing nations," not as an urgent global priority because it's fueling the creation of antibiotic resistant infections which can be exported within 24 hours to their country by someone jumping on a plane before they are symptomatic. And I really can't recall ever seeing a headline explicitly linking the two things. But they are linked. http://resistancecontrol.info/2017/prevention-first-tackling-amr-through-water-sanitation-and-hygiene/ http://resistancecontrol.info/2017/prevention-first-tackling...
  - pkaye 7y ago
    Bill Gates did a lot of funding for developing low cost toilets for developing countries. There is a documentary called "Bill Gates Mind" on Netflix that talks about it.

tjchear 7y ago

I'm not an expert in ML/Biology, but I wonder if doing so won't completely eliminate all of humankind's diseases, but shift the battle from one between humans and bacteria/viruses, to one where ML takes the place of humans by proxy (say we let this ML vs superbugs play out over centuries). I wonder to what direction evolutionary pressure in the face of ML would take bacteria. Perhaps a super smart bug.

ganarajpr 7y ago
Thats a very interesting thought. I am not sure evolution can compete with artificial evolution ( ML ). But we never know :). Fingers crossed that ML can always beat evolution.
- michaelscott 7y ago
  That's also a very interesting thought. The general narrative with general AI is that natural evolution in humans is too slow, we won't be able to catch up mentally with such an AI and this is a bad thing. But the flip side is that natural evolution that is damaging to us (bacteria and viruses) also won't be able to evolve fast enough.
Rochus 7y ago
Molecular biology is not that easy. Selecting a promising molecule is only a small fraction of the work. If you know the molecule you still have to be able to replicate and bring it to the place where it is required. So even if AI can support the process still a lot of plain old biochemistry is needed. AI is just one of many tools. And as you can read in the article a lot of trials are required by the regulatory authorities before the drug can be applied to humans.
- antirez 7y ago
  True but it's interesting that once you have an oracle that given a molecule provides you with an antibiotic score, you can show it directly molecules that are easy to produce and or likely safe for humans. This should help significantly.

throwGuardian 7y ago

What's becoming apparent is the fundamental nature of applied mathematics and ML/AI/DL to our future. Our education system needs to adapt, include this as a mandatory part of a college education (BRIC countries are including this at the high school level). A degree in pure AI/ML without mastery over impactful problems and it's underlying science, is not the ideal future. Every chemist, biologist, ... should be proficient in ML/DL

mantap 7y ago
The field of ML is moving too fast to be taught in schools. It's better to teach its foundations (linear algebra, etc) which are well developed and static.

thedance 7y ago

It's interesting here that the breakthrough seems to come from the computer search not suffering from epistemological block. It doesn't think, therefore it is also not bound by conventional patterns of thought, such as the bias a researcher may show when thinking that this known compound is not an antibiotic, when it is in fact an antibiotic compound.

logifail 7y ago

Decades ago when I worked in a lab, "high-throughput screening" was definitely on the list of buzzwords. Given what I saw back then, I'm struggling to understand how there could possibly be "a library [..] of 6,111 molecules at various stages of investigation for human diseases" (a.k.a. "Drug Repurposing Hub") which hasn't already been partially or fully screened for interesting antibiotic activity. Could it be there's more fame (and funding) in a project where you can publish a paper and get headlines about "ML" and "superbugs", than in actually testing a library of existing compounds to see if any of them kill E. coli?

anonsivalley652 7y ago
For DD, it seems initial screening of as many phages, microbes and compounds as possible using highly-automated brute force might be plenty efficient to test their effectivenesses against every horrible, resistant and opportunistic pathogen for candidate identification. Maybe flying drones out to collect samples in as many random places (public places, restrooms, dirt and even more random places) as possible, generating more samples than a team of humans ever could. There doesn't seem to be any "One True Way," but a holistic synthesis of collection, identification and selection methods.
- logifail 7y ago
  > highly-automated brute force might be plenty efficient Hasn't this been going on in one form or another for many decades? When I was in this field (20+ years ago) I got to visit labs at Glaxo Wellcome, SmithKline Beecham, Zeneca and so on. Even back then they were proudly showing off lab robots which allowed them to run large-scale screening experiments. Not sure any of this stuff is quite as revolutionary as it looks.
  - kokey 7y ago
    I did a stint at GSK about 13 years ago. The large scale mechanised experiments also produced big databases of compounds and their properties and mechanisms to make these accessible to researchers and build bigger clusters of systems to run models on. There was a lot of talk of ML but implementation was nowhere near the scale we see nowadays. I think what has changed significantly is the scale, second to that the methods including algorithms and methods to feed back real data into it. It's an evolutionary improvement but I sense it's one of those things where one lucky step in the evolution could make a big difference in productivity.

tasubotadas 7y ago

Here is the code repository for the model https://github.com/chemprop/chemprop https://github.com/chemprop/chemprop . I am surprised by the (high) quality of the code published there. It is quite a rare case in the academic world :).

phobar 7y ago
They are funded by several big pharma companies (from which they have learned a lot in terms of QSAR). One main point of the collaboration is the creation of production ready toolkits.

loopasam 7y ago

A medicinal chemist take on it: https://blogs.sciencemag.org/pipeline/archives/2020/02/20/machine-learning-for-antibiotics https://blogs.sciencemag.org/pipeline/archives/2020/02/20/ma...

lngnmn1 7y ago

Why, yes, pattern recognition is what ML really is and should be used for. Aside from pattern recognition everything else is a new astrology, with estimated (straight from one's ass) probabilities instead of planets and constellations.

allovernow 7y ago

A lot of the more cynical commenters here are misunderstanding the breakthroughs that have lead to the novel discovery, and underpin the emerging ML revolution that we are just beginning to witness. Yes, partly the results of this study are due to increased compute, but I'd say that's only about 50% of the secret sauce. The other 50% is attributable to many very recent developments in the field ML which are gradually coming together in solutions to a range of problems - deep learning, convolutional networks, new architectures (GAN, transformer, RNN, LSTM, autoencoders, etc), regularization techniques, gradient control, hyperparameter optimization...the list is quite long, and every SOTA neural network inevitably incorporates a large proportion of these small steps towards the massive leaps that are being taken in the applied ML space. The concepts of perceptrons and gradient descent may be 50+ years old, but an entire body of theory has been explosively developed in the last decade, such that comparing modern ML to what was being done just a decade ago is akin to the difference between, say, programming theory now and 100 years ago. And for the same reason - just as compute advances enabled development of computer science, so to have recent hardware advances unlocked a new level of machine learning. But I'd like to point out that, even moreso than software development, very little of the grand breakthroughs we will soon see will be possible without multidisciplinary domain knowledge. It is very difficult to effectively apply ML without a solid technical understanding of the properties of the applied data space, which for real world applications are constrained by physical laws and represented and communicated best by mathematical descriptions. ML engineering is a generalist's game - and what we are going to find is that the most successful ML engineers come from broadly applicable, math heavy backgrounds - physics in particular, electrical engineering, to a lesser degree mathematics, etc - because ultimately training a neural network comes down to adequately sampling a problem space and curating data with an intuition which is most ideally developed by the study of mathematics. It is a very general view of the world which is difficult to communicate to someone who is not experienced with higher math. The current wave of applied ML startups will see a high rate of failure - because ML is still being treated as an extension of programming, in the sense that you expect to be able to hire a bunch of pure developers to translate a specialist's knowledge into code. But this emerging field is different, the few startups that succeed in the applied ML space will be those that are able to find the rare domain experts who have picked up ML along with their math and science experience. There will effectively emerge two classes of ML engineers with substantially different levels of compensation - the coveted generalists who have cross-pollinated with math heavy disciplines, and the rest.

dom96 7y ago

I saw news about this on The Guardian and it was the first time in a while where I was truly amazed by an ML-related discovery. This kind of thing is incredibly exciting and I hope we continue to see similar discoveries in ever more wider fields.

DeonPenny 7y ago

Knew this happen eventually. ML drugs will be huge