AI Creates Killer Drug

Researchers in Canada and the United States have used deep learning to derive an antibiotic that can attack a resistant microbe, acinetobacter baumannii, which can infect wounds and cause pneumonia. According to the BBC, a paper in Nature Chemical Biology describes how the researchers used training data that measured known drugs’ action on the tough bacteria. The learning algorithm then projected the effect of 6,680 compounds with no data on their effectiveness against the germ.

In an hour and a half, the program reduced the list to 240 promising candidates. Testing in the lab found that nine of these were effective and that one, now called abaucin, was extremely potent. While doing lab tests on 240 compounds sounds like a lot of work, it is better than testing nearly 6,700.

Interestingly, the new antibiotic seems only to be effective against the target microbe, which is a plus. It isn’t available for people yet and may not be for some time — drug testing being what it is. However, this is still a great example of how machine learning can augment human brainpower, letting scientists and others focus on what’s really important.

WHO identified acinetobacter baumannii as one of the major superbugs threatening the world, so a weapon against it would be very welcome. You can hope that this technique will drastically cut the time involved in developing new drugs. It also makes you wonder if there are other fields where AI techniques could cull out alternatives quickly, allowing humans to focus on the more promising candidates.

Want to catch up on machine learning algorithms? Google can help. Or dive into an even longer course.

60 thoughts on “AI Creates Killer Drug

    1. When I was doing drug screening on lab bench scale (not a large Pharma company) it was easy enough to just run 96 well plates and you could do a few of those at a time.
      With microarrays and all sorts of stuff now available I wouldn’t call it trivial but it isn’t very “hard”

        1. Was a phone call away. Combinatorial libraries are widely available. There are also research institutes with their own 10’s of thousands of compound libraries that, with grant money, will run your protocol /assay and report back results. At the time we assumed that even that was a drop in the bucket compared to what happens behind the closed doors of Big Pharma. High Throughput Screening was the jargon of the day.

          1. OTOH the machine learning saved all that time and costs in both compounds, delivery and actual work. This means that smaller companies that are agile enough can save ton of time and money on early research and use this advantage later. I’m pretty sure that in Big Pharma companies a meeting with management to tell the about such research method would take more time then research itself.

            Consider this: IBM developed PC because someone decided to create a team working on it basically outside the organization and its culture. They had a hit product, so when IBM PC started to sell, that team was taken back into the fold. And managed. In the end they lost to smaller companies that made PC clones cheaper and faster than they could.

            Consider this: a piece of software can test tens of thousands of new compounds in matter of hours. If it has a model of the problem and some training data, it can select or even develop new compounds for basically any problem. So today it can search for antibiotic to eliminate a drug-resistant bacteria, tomorrow it can develop a new cure for specific type of cancer. Next week – regrowing amputated limbs. Or more profitable research into fat loss drugs and hair growing for bald people.

          2. Are combinatorial libraries actual chemical compounds?
            How easy, efficient and affordable are research institutes to charge them with testing thousands of chemicals? Research institutes always present themselves as…. research institutes, not as contract labs. So, just to find the right institutes looks difficult.

    2. A word of caution: creative people usually make things for the good of humanity. But not always. The same AI can (and probably will at some stage) be used for nefarious and possibly dangerous purposes – and I expect it will be equally effective. Are we ready enough for that outcome?

    3. More than merely testing of the compounds, it would require synthesizing each first, many of which would be very expensive or impossible to produce for testing.

        1. I’m reminded of the old joke, “If hooking a car battery to a monkey’s nuts would cure cancer all I’d have to say is ‘Red is Positive and Black is Negative’.” While one day we will be able to do this research without animals today is not that day and I still value a human life above that of a rat, I apologize if that offends anyone.

        2. Agreed. Even though animal testing is done as humanely as possible, it still requires one swallow hard and make something of a peace with him or her self. Regarding testing on humans: I seem to recall the last group of people who did that in earnest are not remembered fondly.

          1. I hate to tell you this but human testing occurs on EVERY DRUG before it is approved by the FDA. And many are rejected because of the severe reactions and side effects they cause. All of the subjects are now (hopefully) volunteers. It’s all part of the process! As for animal testing, any day and twice on Sunday if it saves a single human life or betters our existence. Sorry, that’s just how I feel.

      1. I really hope that some day (the sooner the better!) all this AI and stuff will put an end to torturing animals “for science/research/whatever”. No problem for messing with stuff in vitro, but please stop torturing animals!!!

        Also i won’t look for details but i know they were several molecules that had been “tested” on animals and then on humans and it went horribly wrong, ie the molecules where found to be highly toxic/dangerous to humans only.

        1. You know that researchers can’t just walk up to the research animal store and be like “yeah, I’ll take a dozen gorillas cause I want to see if they go blind if I inject their eyes with ink,” right? There are so many hoops that they need to jump through, ethics reviews that need to be satisfied that the study must be conducted on animals instead of computer models, that the experiments can’t be adequately done on lower order animals, like invertebrates, and that the animals that are approved for use are minimal in number and that they, given the circumstances, will be treated as humanely as possible, including the point of sacrifice.
          Animal testing is a necessary part of advancing medical science and anyone who uses any medical drug is inherently complicit in the use of animals for the benefit of human life. Making peace with this will not only alleviate you of your moral high horsing, but also let you understand that the world is a nuanced and complicated place with people generally trying their best to do the right thing, even if how they’re trying isn’t something that meets your approval.

      2. While I agree in spirit that I wish there was another way other than animals, the standard for animal testing is that nothing more painful than a needle stick is permitted. And the animal facilities are highly, HIGHLY regulated and monitored for compliance. The animals are very well cared for until, you know, they are euthanized. It’s also overseen by a team of care professionals and at least one veterinarian.
        I’d say they are more kindly treated than food animals so until we get rid of all meat for human consumption (it could happen) animal testing is here to stay.
        And, as always, before all the “gotcha!” stories, I know the history of animal (and human!!) testing and the missteps and f-ups along the way. But overall what I said holds.

    1. Doesn’t matter if you can inexpensively test the result. If testing the result is cheap enough, literal randomness can be a way to invent solutions. It absolutely does not matter how a system works, IF the results can be tested and verified.

        1. That would be an appropriate criterion for a process, sure, but once a drug has been through the testing and clinical trial process (this hasn’t yet) it doesn’t matter if it was invented by a pharmacologist, a soil bacterium, or an inscrutable connectionist algorithm.

  1. “killer drug” eh? I expected this to be a story about AI gone wrong and the drug being
    toxic to the user. Well, in time we may see things develop in that direction.

    1. Although I wasn’t really hoping for that be the case, I feel the title is misleading. Especially after Sam Altman testified in Congress. Still an interesting article though…

      1. Lol funniest things I’ve read. Lol I’m so tired of the same old meth and heroin. When will a computer finally create something new. Lazy drug manufacturers, idk when the last time they made a new drug was.

        1. Not that anyone, but the so called ‘elite’ of the world, would ever be privy to any of that for themselves.. we are merely the fodder that feeds their lifestyles after all ..

  2. Yeah it’s nice and totally amazing that AI can assist in the creation of a drug that specifically targets a disease without harm to that diseases surroundings, until you ponder the obvious question, ” what if AI considered us to be the infection”.

    1. Your gross misuse of the word killer is completely out of context and is incredibly misleading.

      I get that you’re utilizing shock factor to increase traffic to the article, not to mention encouraging fear-mongering.

      Without those, this is just a plain, biased informational article. Hmm, do better.

  3. The program reduced the molecules from 6700 to 240. I guess it’s not really science until they test all the molecules to see if the AI missed any good candidates. It could have just randomly picked 240. Do you think the scientists added any molecules that wouldn’t have worked to test the AI. “Sir, the AI picked mustard and ranch as potential antibiotics”.

  4. deep learning could also be used to develop very potent drug to kill humans. almost certainly already in development, given all the other known active death technology programs.

  5. {It isn’t available for people yet and may not be for some time — drug testing being what it is.}
    Tell that to the researchers pushing the Covid vaccine active testing on humans still to this day….

  6. Now I want to see what AI comes up with for the most potent nerve agents. I mean we have fentanyl which is absurdly potent to morphine, what compound is absurdly potent to vx? I wonder what one can do with a chemical so wildly toxic that I can use it to take over the world.. see? AI is the perfect risk for a species like us humans. AI is likely the reason why most civilizations stop advancing to the point of interstellar travel.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.