15 comments

  • entee 1498 days 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 ;)

    • xzel 1498 days 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 1498 days 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 1497 days 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 1498 days 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.

        • m3kw9 1497 days ago
          Problem with many of these molecules that are found to work in mice, when put in human which is considerably more complex, say it goes thru the liver, the liver may break down most of it when circulating the blood stream. So now the drug needs to also resist that type of break down. Now multiply that by 10 to 100 of other types of reactions your body can generate from this.
          • pacman128 1497 days ago
            Are humans more complex than mice or just different? I've always assumed the latter, but I'm in no way an expert.
            • whatshisface 1497 days ago
              The mouse genome is about 14% shorter than the human genome, although that could mean a lot or a little depending on certain other factors.
          • sjg007 1496 days ago
            Well c diff is in the gut so it’ll work there probably.
      • conjectures 1497 days ago
        > People have been doing this exact thing for two decades at least but obviously with less computing power.

        Here's an analogy which shows how the cynicism here isn't actually useful wisdom:

        Lee Sedol lost at go.

        Kasparov lost at chess two decades before.

        The Kasparov loss was 'exactly the same thing albeit with less computing power'

        • SQueeeeeL 1497 days ago
          Correct. I don't understand why this contradicts the argument.

          "Computers can learn complicated games with more computing power"

          "Compute power goes up"

          "Computer learns more complicated game"

          I mean, if you really think of AlphaGo on an extremely high level, it's just a really elaborate way to create and learn a dictionary of moves to take under different circumstances. Of course that's going to be completely dependent on amount of memory and CPU power.

          • Bartweiss 1494 days ago
            I think the criticism is that it's not obvious whether success here was a function of improved performance, expanded throughput, expanded testing, or sheer luck.

            Chess engines have clearly improved in both design and computing power over the years; doubling an engine's resources or pitting a new engine against an old one produces straightforwardly better play. But the drug-discovery technique in use here may not be "playing better" in terms of producing higher-quality predictions.

            To extend the chess metaphor:

            - Deep Fritz is a stronger player Deep Blue even with 4% as much computing power. This story does not appear to be an algorithmic breakthrough of that source.

            - Deep Blue lost to Kasparov in 1996, then beat him in 1997 with double the computing power. That's a clear improvement in play, but not an improvement in efficiency. This story might represent such a change, modelling more prospective drugs to test higher-confidence candidates.

            - If an AI that can only win 2% of games against humans plays 10 games, it has an 18% chance of beating someone. But over 100 games, it has an 87% chance of a win. This result might be a team with a larger testing budget claiming the 'first win' without any AI-side improvement.

            - If a dozen grandmaster-level chess AIs play GMs, one of them will have to get the first win against a human. Labeling this result a 'breakthrough' in AI terms might be outright publication bias among equivalent projects.

            As far as the drug, none of that really matters, except that efficiency improvements would have more potential to increase drug discovery. The drug itself is still useful, and the discovery is a proof of concept; in 1980 no possible computer would have beaten Kasparov. But this is being hailed as a breakthrough in AI in seriously questionable ways. The BBC article, for example, managed to imply that this specific project was novel and important for using neutral nets to produce a significant result.

      • Psyladine 1497 days ago
        It's a bummer, from the headline you'd think the ML designed a molecule and tests with it showed results, instead it compared known drug interactions against available compounds, sort of a rainbow table matrix. You know, what researchers already do for a living.
      • Dumblydorr 1497 days ago
        There are differences over time, though. This drug repurposing, for instance, allows us to take drug candidates for other diseases and apply them to new areas. That is a fairly new approach when combined with ML in drug development, as in the past one would see brute force computation without repurposing. Repurposing gets better and better over time, as well, since there are more and more drugs on the lists that we've already run through a gamut of testing.
      • redsymbol 1497 days ago
        > 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 guess I'm not sure where the dismissiveness is coming from here. Are claiming this could have been trivially done before? If so, why didn't you or someone else do it already?

        Or are you claiming it's an uninteresting result that is not worthy of publication or attention?

        • Bartweiss 1494 days ago
          This is a novel and important result in antibiotics. It's also a proof-of-concept for using ML to produce vital drugs with novel mechanisms, rather than incidental alterations or discoveries in noncompetitive spaces. It might be an incremental speedup or computing-power advance in ML drug discovery also, but it could equally just be the result of a lucky break or a particularly large lab-test budget. (In which case, "why didn't someone do it already?" is closer to asking why nobody else bothered to win the lottery.)

          It's not a major theoretical advance in ML drug-discovery techniques or the first big step in ML drug discovery. It's certainly not the invention of ML drug discovery or neural nets as an ML technique, both things I've seen implied in news stories on this work.

          This is attention-worthy, absolutely. (I'll leave "publication-worthy methodology" to experts.) But it's newsworthy on actual merits, as a drug breakthrough and a demonstration of an increasingly-important technique. So I share the frustration when lazy or confused reporting implies this is the same style of ML-theory breakthrough as CNNs, Transformers, or even neural nets themselves.

        • xzel 1497 days ago
          I'm saying I'm tired of the puff piece articles and reddit style headlines. I was saying NN's and other "AI" style models has been used in this way for decades for these types of things. 0 disrespect to the actual science; the paper is great and I truly hope it works in humans. We need more drugs against the inevitable fight against drug resistance.
          • redsymbol 1497 days ago
            Okay, fine... what do you want to be different about the situation, then?

            Do you not want any non-technical summary articles like this to be written? So that only those with the training to understand a Cell journal article would be able to learn anything about the result?

            Or do you prefer that no journal articles be published that rely on 2020-era NN models, because older articles based on less state-of-the-art NNs have been published already?

            • xzel 1497 days ago
              I'd much prefer there be more detail in the article about AI research in drug discovery, how it still needs clinical trials in humans. I agree with the sub-comment as well. I think your last question doesn't make sense to me though as I haven't said anything similar to that.

              But I think you raise an interesting point about non-technical summary articles: what do they do and who are they for? Does the non-technical public need to know about this research? Do they gain anything vs. reading the study's actual summary? I'm not sure. I do think there isn't much for non-technical people to get from this article that would be useful. I think the best reason would be for younger people to pique their interest in the field. Though, I honestly don't know the answer.

              • redsymbol 1497 days ago
                Yeah, good points. I guess such non-tech articles serve several purposes.

                I do think this one gives non-technical people get something useful, though. More broadly, it'll increase science understanding among the non-scientist/non-technical public, at least on this topic, and good does tend to come out of that.

            • sp332 1497 days ago
              The story could be about perseverance paying off, instead of a breakthrough portending big changes in near-future drug development.
      • asdfman123 1497 days ago
        Yeah, I used linear regression through Excel in my 5th grade science fair to draw a nice line over my data points that showed how plants grew with different fertilizers.

        Linear regression is everywhere in science and that can be classified as machine learning artificial intelligence.

        Still, it's cool that science is using more data and more intricate algorithms to produce fits.

      • mattmanser 1498 days ago
        I don't understand, they didn't find ONE they found NINE.

        That doesn't sound like luck.

        • xzel 1497 days ago
          That is sort of the thing about receptors, it is really had to know what will actually work. When you run these types of programs you'll get a number of possibilities that the simulator will spit out, that be from an AI or more traditional stat generated model. There were a few articles very similar to this with Coronavirus cures. They then have to go work in a well designed clinical trial, against a placebo, and then in humans, which is way more unlikely. It's most likely that all 9 don't work, unfortunately. I'm not saying the fact they found them was luck, the luck is if they actually work.
        • abrax3141 1497 days ago
          Finding nine is more like luck then finding one. Note that just one of the nine (so far) is validated. If I tell you to roll a six on a dice and you do that’s pretty lucky. If I tell you to roll an even number (I.e., give you wider scope) and you do ... less amazing. They could have lab tested every compound in the set but it would have taken too long. The ML for them to 9. I’m not saying that this isn’t good work. Just addressing the statistical point re 1 v 9.
    • chrisgd 1498 days 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 1498 days 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 1498 days 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 1498 days ago
        They do. It's just easier to add extra CPU cores to a ML pipeline than another person.
      • IAmEveryone 1498 days 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 1498 days 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.
    • naresh_xai 1498 days ago
      Exactly why we need causal reasoning/causal proof alongwith ML
      • entee 1498 days 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 1498 days 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.

  • thedance 1498 days 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.
  • koolba 1498 days 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.

    • kirrent 1498 days 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.
      • whatshisface 1497 days ago
        Here's an interesting thought. Let's say that the number of possible antibiotics that can be discovered is so high that bacteria cannot simultaneously be resistant to all of them (plant immune systems, which involve a lot of small moleculea, indicate that this may be possible). Then, this game theory trap where stewards are guarding what we have so that pharma doesn't want to make anything new is both pointless and self-perpeuating, because we will never reach the point where we realize we've beaten resistance.
      • SeanAppleby 1497 days ago
        Yeah, this seems game theoretically problematic. We clearly want a deep warchest of antibiotics, but new antibiotics won't sell in any meaningful quantity likely, or even hopefully, until after their IP rights expire. I don't see how you make money here.

        Maybe the government should structure some new incentives for stocking the antibiotic warchest.

      • rkangel 1498 days 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.
    • westurner 1498 days 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

      > Mutant generation

      > Chemprop code is available at: 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/

      > 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.

    • Dumblydorr 1497 days ago
      The issue is that no pharma company wants to make drugs that will be carefully stewarded as last lines of defense vs super bugs. You won't see 3M patients per year because every hospital system worth its salt has antimicrobial stewardship programs which will interrogate every case of antibiotic use. Doctors are trained and pushed into using these drugs as infrequently as possible because they all know Darwinian selective pressure will render antibiotics less useful the more they're used.

      But it's a different story in developing nations with lower standards, where they will steal your IP and bombard the microbial population with every antibiotic they can.

      Furthermore, agriculture will surely love to grab new compounds to blanket their herds with. They will not give pharma great profits either, and they'll increase resistance to your fancy new drug that you spent billions developing.

      So overall, it's a grim financial picture for pharma and antibiotics. They would prefer to make cancer drugs, chronic disease drugs, biologics, and not these carefully stewarded and easily ripped off antibiotics. IMO, antibiotics will come from altruistic scientists working on their own, and not from the CFOs and bottom lines of pharma.

    • loopasam 1498 days ago
      A couple of good recent article to understand this complex problem (USA perspective): https://endpts.com/can-we-make-the-antibiotic-market-great-a... and https://endpts.com/biopharma-has-abandoned-antibiotic-develo...

      Reasons are mainly: No incentive to develop antibiotics from a legal perspective (FDA), as insurance companies prefer to reimburse the cheap and generic, still working mostly "well enough" for now.

      Insurers pay for in-patient antibiotics as part of a lump sum to hospitals known as a Diagnosis Related Group (DRG). Using a cheap antibiotic increases hospital profit margins, while using an expensive new drug could mean that a hospital might lose money by treating a given patient. As a result, hospitals are incentivized to use cheaper antibiotics whenever possible. This puts significant pricing pressure on new antibiotics, which are one of the only type of medicines paid for like this.

    • cbhl 1498 days ago
      Here is a prior HN submission that talks about the economics of new drug development for rare diseases:

      https://news.ycombinator.com/item?id=19787367

      tl;dr: a hypothetical drug with an expected profit of $547 million, adjusted for discounting and risk, has an expected value of negative $6 million.

    • refurb 1498 days ago
      In terms of your numbers, 2.8M is all drug resistant infections - even the best drug would only work for a fraction of those..
  • tjchear 1498 days 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.

    • Rochus 1497 days 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 1497 days 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.
        • Rochus 1497 days ago
          It will not go without replication, trials and production, regardless what oracle you have. And keep in mind that the oracle has a quite significant error rate. You won't get an immediate solution, just proposals. And whether or how a drug works is also dependent on the environment and there are dynamic aspects and complex processes within a cell extending over space and time. It's much more complex than speech recognition or self-driving cars.
          • tjchear 1497 days ago
            We speak in hypotheticals of course, but barring any physics defying barriers here, one can imagine a future where those issues you raised have been addressed (e.g improved simulation/ML techniques, less red tape, faster trials, greater accessibility to treatment). Even now, I think mankind's immune defense is no longer confined to itself, but must be thought of as a bigger system that extends to the pharmaceutical industrial complex. Any sufficiently ambitious bacteria is basically taking on not just us, but also the collective work of all researchers diligently working on the next super bug killer.
            • Rochus 1497 days ago
              You are invited to try it. You will not be able to avoid the laws of nature, and there will also be many legal hurdles. But if you make a big step forward, you will undoubtedly be a Nobel Prize candidate.
    • ganarajpr 1498 days 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 1498 days 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.
  • loopasam 1498 days ago
  • logifail 1498 days 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 1498 days 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 1498 days 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 1498 days 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.
          • logifail 1498 days ago
            > one lucky step in the evolution could make a big difference in productivity

            There's also the issue of ROI. Is Big Pharma really expecting to find an antibiotic blockbuster drug?

            20+ years ago there was a distinct lack of excitement around antibiotics in general, at least from the commercial types.

            Q: Is there more expectation/excitement/R&D budget now?

        • anonsivalley652 1498 days ago
          I think you may have missed my point. Human labor of 10 workers costing $ 1+ million USD in total OpEx isn't as scalable as sending out 10,000 drones to collect samples. Drones that can return in 45 minutes, don't need lunch breaks, and don't have to find parking.

          Industrial-scale sample processing -> putting more people a little farther down the pipeline to actually look at what's interesting rather than doing unnecessary/low-skill field work.

  • throwGuardian 1498 days 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 1498 days 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.
  • antipaul 1498 days ago
    Thank goodness it was "aided by machine learning". Otherwise, would anybody be reading this article ;)?
    • wyattpeak 1498 days 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 1498 days 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 1497 days 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!
  • tasubotadas 1498 days ago
    Here is the code repository for the model 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 1498 days 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.
  • dom96 1498 days 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 1497 days ago
    Knew this happen eventually. ML drugs will be huge
  • allovernow 1498 days 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.

    • superpermutat0r 1498 days ago
      I think multidisciplinary knowledge is unnecessary, especially now with deep learning where there is no feature vector design.

      Even the DeepMind discoveries in various fields always feature the same people that I doubt know deeply about protein folding or similar stuff.

      Even when you look at computer vision research and how all the sophisticated methods became unnecessary when NNs came to dominate shows the same thing.

      I remember having to learn about dependency parsing, part-of-speech tagging, named entity recognition, entity relationship inference, document summarization and a bunch of sophisticated modelling. Combining all of that to get to high level tasks like machine translation or question answering or even summarization (some methods pruned the dependency tree to get a summarized sentence) was difficult.

      Look at transformers disrupting the NLP. There is no concept of dependency tree, no need to do POS tagging, it's not even necessary to think about that when making a machine translation system. People were figuring out how to build better and faster dependency parsers, POS taggers etc. Domain knowledge was massive and it became redundant with the advent of transformers.

  • lngnmn1 1498 days 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.

  • ErotemeObelus 1498 days ago
    This method might be generalizable to prions.
  • DoreenMichele 1498 days 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.

    • throwGuardian 1498 days ago
      Super bugs are a mostly first world problem, rampant in top tier hospitals providing cutting edge treatments, despite following safety precautions meant to deter secondary bacterial infections during or post hospitalization
      • refurb 1498 days ago
        Super bugs are a mostly first world problem

        Not true at all. There are several known resistant strains that have come out of developing countries.

        Why? Antibiotic use can be rampant - in many countries you can buy them without a prescription.

        • mnw21cam 1498 days ago
          I heard someone (in the correct area of expertise) describing how he had visited a developing country and tested the water in a river downstream of an antibiotic manufacturing plant, and found the concentration of a particular antibiotic to be approximately the same as what you would typically want to achieve in the blood of a patient.

          If that isn't going to drive resistance to that particular antibiotic, I don't know what will.

          • perl4ever 1495 days ago
            People say things like this, but how many antibiotic compounds are in the soil naturally? Living creatures exist in an environment with countless bacteria and evolve to survive, and some of the compounds they produce are turned into human drugs. So I feel like there's something missing in the ordinary narrative about antibiotic resistance. Where did penicillin come from again?
      • DoreenMichele 1498 days ago
        Most developing countries don't actually keep adequate data on super bugs. So we don't really know that.

        In 2014, the WHO released the first report on antimicrobial resistance, in which the WHO collected national data on nine bacterial infections/antibiotic combinations of greatest concern for global health.[5] The data revealed that out of 194 countries, only 129 provided data, of which only 22 had data on all nine infection-antibiotic resistance combinations deemed to be emerging global threats.

        https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6380099/

        Anyway, this is sort of off topic because I know the article was submitted to HN for the cool factor of using artificial intelligence or machine learning or whatever, not because people are really deeply concerned about antibiotic resistant infections. So idly saying "I feel like they are barking up the wrong tree and focusing on the wrong things if they want to solve this." isn't at all welcome as part of the conversation, as evidenced by the downvotes for a really mild, brief remark.

        I wasn't trying to derail the discussion. I just read up some on things like antibiotic resistance because of my medical situation, so I happen to know a bit about the topic.

        But I'm planning on stepping away from this discussion because it seems like a pointless waste of time that's just inviting negativity from a crowd that doesn't see my comments as anything intelligent or interesting.

    • bravoetch 1498 days ago
      We do.
      • DoreenMichele 1498 days 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-...

        Price to pay: Antibiotic-resistant infections cost $2 billion a 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...

        • ranDOMscripts 1498 days ago
          That's probably because Bill & Melinda issued their grants a few years back. You can see a list of all of the high-tech outhouse makers here [0]. Many of the designs are out in the field for long-term testing.

          [0]https://stepsforsanitation.org/innovation-center/

          • mikorym 1498 days ago
            I come from a country with serious sewage issues. Most of it is government operated sewage works that are just not maintained and were due for upgrade 10-20 years ago. The issue is also an order of magnitude worse in some areas than in others.

            The Gates Foundation is not relevant here, in that problem field, in any meaningful way. It is an issue about operational structure and unit cost structure. Tax money gets spent on black Mercedes cars instead of what it is supposed to be spent on.

            Also, it is not a technological issue at all. In terms of tech for sewage treatment, the path forward is clear. In terms of politics it is not.

            • ranDOMscripts 1498 days ago
              It is my (albeit limited) understanding that the Gates grants are focused on providing toilets to areas where there previously were none and not on replacing crumbling infrastructure. They're trying to get facilities to the millions of people who have to poop in the bushes because that's all there is.
              • mikorym 1497 days ago
                That's not the issue. The issue is that there are sewage systems that dump raw sewage into rivers. Toilets are more of a nod to human decency than environmental issues.

                But if Bill would like to phone me, then I would explain to him that while decency does score high in my books, you can have perfect golden toilets that dump sewage into rivers. In that case I would rather take a veldtie in the bushes.

          • DoreenMichele 1498 days ago
            Or maybe it's because you don't see articles about open defecation posted here all that much.

            I can find two. One from two years ago and one from six years ago.

            https://hn.algolia.com/?q=open+defecation

            If you search on toilet, there's a lot more articles that come up, but at first glance, most don't appear to be about solving open defecation, though there is one on the front page of the search about the Gates foundation funding toilet research.

            https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...

            In contrast, you can put in antibiotic resistant and limit it to the past year and get pages of hits. Granted, they won't all be about the latest sexy research methods.

            https://hn.algolia.com/?dateRange=pastYear&page=0&prefix=tru...

            • eesmith 1498 days ago
              There's one more. https://news.ycombinator.com/item?id=20730146 from six months ago is titled "California’s Biggest Cities Confront a ‘Defecation Crisis’" and deals with "excrement on the sidewalks of San Francisco".

              It has 159 comments, probably because of how it affects "first world wealthy people".

        • pkaye 1498 days 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.
        • lazyasciiart 1498 days ago
          I think those links look interesting, can't speak for the rest of HN. If you don't want to post them, can I? (specifically that last one?)