AI's Medical Marvel: Uncovering the Antibiotic Combatting Deadly Superbugs

0:00Today on the podcast, we are going to be talking about scientists using AI to help discover new antibiotics and to treat deadly superb bugs. One case in particular that's interesting, and then we're going to be diving into what the implications are for science, for drug discovery in general, which is a multi-billion dollar industry, and what the impacts are on humanity as a whole from this. So the thing that we wanted to talk about and kind of the headline news is that recently scientists using AI discovered a brand new antibiotic that can kill a deadly superbug. So according to a new study that was actually published just on Thursday in the science journal Nature Chemical Biology, just a whole bunch of scientists from McMaster University and also MIT, they discovered a new antibiotic.

0:47So the superbug in question is called Acinatobacter bomanii. And the World Health Organization actually classified this as like as a critical threat among they kind of have like a priority pathogens group of bacteria that they say pose the greatest threat to human health. And looking into it a little bit, honestly, it was a little bit terrifying what this pathogen can do. And so I understand why it was classified as such a high threat. and previously they did not have a solution for this. They didn't have something that could actually help this and one in four people that contracted this within a month actually died from it.

1:29So this superbug was really deadly. It was a big problem. They didn't know how to treat it and it had a lot of different reasons why it was so deadly. But according to the World Health Organization, the bacteria has a built-in ability to find new ways to resist treatment and it can pass along genetic material that allows other bacteria to become drug resistant as well so basically from my understanding it actually can collect DNA from things that it comes in contact with and it can find what in that DNA is resistant to its you know its to its lethality or whatever and it can it can kind of optimize and overcome those so that it's so that you know if you're using different antibiotics for example it can optimize to overcome those antibiotics and still be a lethal pathogen which is I don't know in my opinion a little bit terrifying it specifically poses a threat to like hospitals nursing homes and patients who require ventilators and blood catheters as well as people that have open wounds from surgery are those are the people that are most at risk from this.

2:38And it can live, this is what I don't know, this kind of what scared me the most about it, it can live for prolonged periods of time on like surfaces and shared equipment. So if it's, you know, on like a hospital bed, or if it's on the table, and something wasn't sanitized correctly, right, obviously, the hospitals, they tried to sanitize everything the best they can. But sometimes something might have been missed. And in that case, it can live for a prolonged period of time on that until it finds a new host. So according to the CDC, it can also colonize or live in a patient without causing infection or symptoms.

3:12Okay, so enough about talking about why this pathogen is terrible. I think we all get the picture. It is a really deadly pathogen. I wanted to highlight all of that though, just to say how impressive this actually is. So the study that they did, they used an AI algorithm to screen thousands of antibacterial molecules in an attempt to predict new structural classes. So as a result of the AI screening, the researchers were actually able to find this entirely new antibacterial compound, which they named Abusin. And so one of the researchers, Gary Liu, he's a graduate student from McMaster University, and he worked on the research for this, but this is what he had to say about it.

3:54He said, we had a whole bunch of data that was just telling us about which chemicals were able to kill a bunch of bacteria and which ones weren't. So his job was to train the model, and all that this model is going to do is tell them essentially if a new molecule will have antibacterial properties or not. Then, basically through that, they're able to just increase the efficacy of the drug discovery pipeline and hone in on the molecules that they really wanted to care about. So after they trained this AI model, they then used it to analyze 6 ,680 different compounds that it had previously not encountered, right?

4:34So they trained it on one set of data, and then they introduced a brand new set of data, which is important. And I do want to highlight like this is really powerful because if they wanted to like, let's say, go to a lab and actually go and test like 6 ,600 actual laboratory testing experiments, that would take a phenomenal amount of time. And so when they're doing these kind of testing, using an AI model that they can just feed data into it and have it automatically trained and spit out, you know, what the results of that situation or the combination of that compound would be is really, really impressive.

5:11And it really speeds up the drug discovery, right? Like, let's say you could get five of these or 10 of these done a day in a lab. Now you're doing 6 ,600. Like, it's just astronomical the speed that this speeds up. So the analysts took about an hour and a half. Come on, that's incredible. 6 ,600 compounds analyzed in an hour and a half. And it ended up producing several hundred compounds. So it gave them about 240, which they then went and tested in the laboratory. So they went and did all the, you know, the legwork, but doing 240 in a lab is a lot faster than doing 6 ,600. So the lab testing then showed nine that had much higher potential as antibiotics, and Abuchin was one of them.

5:55And then the scientists then went and tested the new molecule against this one particular pathogen, which was Bumani, in a wound-infected mouse. And they found that the molecule suppressed the infection. So they said that the work validates the benefits of machine learning in the search for new antibiotics. And using AI, they can rapidly explore vast regions of chemical space, significantly increasing the chances of discovering fundamentally new antibacterial molecules. they also said we know broad spectrum antibiotics are suboptimal and that pathogens have the ability to evolve and adjust to pretty much every trick they throw at it so ai methods help give them the opportunity to really quickly increase the rate that they are discovering new antibiotics and they can also do this at like a reduced a much reduced cost right like being able to throw in 6 ,000 and have it spit out 200 and you only have to go test 200 really cuts down on time and cost.

7:02So I think this is a really important avenue of exploring for new pathogens. And something else that I thought was interesting about this is the researchers also tested this against about 41 different strains of antibiotic resistant acinine bambamone or whatever like that drug, there's like 41 different strains of it that they tested the same thing against. And it worked on all of them. Although they said it's going to need a little bit further refining to test it in human clinical trials before they can actually use it on patients. Obviously, they'll probably have to get FDA approval and different things.

7:41But while they're working on that, it is really, really promising. And I think what's even more promising is that the compound identified by AI, it worked in a way that stimied only the problem pathogen. So it didn't actually seem to kill many other species of beneficial bacteria that live in your gut or on your skin, for example. And so it's actually a really rare, narrowly targeted agent. So that's something that people are saying is really impressive because a lot of times with antibiotics, if you've taken them before, they kind of just kill all bacteria. And so like a lot of times people can get all sorts of other problems when they take antibiotics like yeast infections or other problems because it's killing bacteria inside the body that is good bacteria.

8:29Where this is really impressive because the AI was able to identify a bacteria that specifically only targeted and killed that pathogen and all 41 variants of that pathogen. So very specific and very effective. And so yeah, the researchers essentially just said that it can prevent bacteria from becoming resistant in the first place, right? Right. So someone that didn't work directly on this study, but kind of oversaw what they were or looked at what they were doing and is doing other things in the same area said that it's incredibly promising. They also said that this type of approach to finding new drugs is an emerging field that researchers have been testing since about 2018 and that it dramatically cuts the time it takes to sort through thousands of promising compounds like we looked at.

9:15right over 6 000 um and so this one particular doctor de la fuente said i think ai as we've seen can be applied successfully to many domains and i think drug discovery is sort of the next frontier um so for this specific study they focused on that one bacteria um because it's in hospitals and a lot of uh settings like that um and they said it's what we call in the laboratory a laboratory, a professional pathogen. So one of the researchers that worked on it was commenting and that's what they said. And then the other thing that I did want to add is that they said it appears to work in a completely new way by preventing components of the bacteria from traveling from inside the cell to its surface.

10:04So what's interesting about this is it isn't just a antibiotic that, you know, addresses the strain, but it works in a new way that might not be very common. A lot of people might not know about, and it might not be something that we are researching for. It's really hard to know like what you know and what you don't know. It's really hard to know what you should research for, right? If there's 6 ,000 possible pathogen antibiotic solutions, and you know let's say like 5 ,000 of them are ones that we would typically we like we understand how it addresses the bacteria we might study those first whereas the AI is able to identify and prioritize ones that we may not even understand how it works so specifically he said that it's a rather interesting mechanism and it's one that is not observed amongst clinical antibiotics so far as he knows.

10:56So this isn't even something that we're seeing in normal clinical antibiotics. The AI like essentially discovered something that works for a reason that is not common or that we it's like it's discovered kind of like new science or a new way of doing something, which is what I think is so incredible about this and about AI models in general and large and neural networks in general is they're able to think of things in new ways that we don't think of. Like there's so many different biases just because of like the way in all areas, but like, let's talk about medicine, right? Like in the medical school or the medical field or in, you know, biology, like you learn about the ways we've discovered how to do things in the past, but we don't know what we don't know.

11:38And there's so many obviously new ways that we could and should be doing things. And so it's really cool when you can take an AI and a model like this and feed in so much more data than a human or a lab could ever get done in a reasonable amount of time and have it test out so many different new areas that then we can go and test right it's not like the ai just like discovers the magical thing it just gives us a lot of really good suggestions helps us narrow it down and then humans obviously go in and test and find out what's actually working so in addition to that he said with this molecule because it only works very uh potently against acetin basur it doesn't impose that universal selective pressure.

12:20So it's not going to spread resistance quite as quickly, which like, as we know, pathogens are always evolving, which is not great. But it appears that this new one they discovered is going to help slow down that resistance or stop it. And so really, really amazing discovery by AI in general. And I think this is going to be a really massive area that people are going to be looking at in the future, antibiotics, but also all sorts of drug discovery. If we could really cut down a lot of the costs associated with drug discovery, perhaps we'd be able to cut down the costs associated with, you know, getting drugs and all sorts of things.

12:57So a lot of possibilities open up. And I think a lot of, you know, the ability for medicine to be less expensive, although, you know, I guess that's still left to be seen um you know if uh some company discovers a drug that they could still make a billion dollars off of even if they were able to discover and uh you know get approved for much less than it costs in the past they may still try to like get what was previous market value for it but i feel like inevitably uh prices will would probably drop um because for example in this study they identified nine different compounds that address this specific pathogen and i think they went with the one that perhaps worked the best or they liked the most.

13:37But if there's nine, I like to think that that would give, you know, other companies nine other options for different possible solutions, right? Like with the, yeah, with all sorts of advancements in medicine, I feel like we see that. So I would like to think this is in, this is a way to not only make a drug discovery cheaper and more effective, but also cheaper for the consumer. So that will be an area that I think we're going to have to follow, but all in all, a really incredible feat for science and all thanks to AI, which is an area we'll continue to follow.