New drugs enabled by artificial intelligence

00:00:04: If you could hear AI, it might sound like this.

00:00:09: Incredibly large amounts of data are processed in a matter of seconds.

00:00:13: A flood of information that is unimaginable to us is analyzed, new information is created.

00:00:19: This information is stored and included in the next analysis.

00:00:23: The network grows and grows

00:00:26: and grows.

00:00:28: What still sounds somewhat futuristic has long since become part of our everyday lives and is now indispensable.

00:00:35: There are problems.

00:00:37: It is no longer easy to distinguish between true and false.

00:00:40: But there are also opportunities.

00:00:43: Not only for making our everyday lives easier, but also for medicine.

00:00:47: But first things first.

00:00:49: Biology and computer science may sound contradictory at first, but computer science is a bit like a manager in a large company for modern biology.

00:00:58: After all, where large amounts of data are generated from research, Biological and medical research is difficult without computer science.

00:01:06: That's why bioinformatician Professor Andreas Keller uses AI in his clinical bioinformatics research group at the Helmholtz Institute for Pharmaceutical Research, Saarland-Hipps.

00:01:18: It was founded in two thousand nine by the Helmholtz Center for Infection Research and Saarland University.

00:01:24: In this episode of Infect, host Julia Dehmann talks to Andreas Keller about how he uses artificial intelligence to understand how beneficial and harmful bacteria communicate in our bodies, to find out when infections cause long-term effects and to research new drugs against pathogens.

00:01:42: He works closely with Saarland University Hospital, which means he's actually very close to people again, while at the same time working with huge, growing data networks.

00:01:55: I'm talking today to Professor Andreas Keller, head of the research group Clinical Bioinformatics at the Helmholtz Institute for Pharmaceutical Research Saarland.

00:02:06: Hi Andreas.

00:02:07: Hey, good morning.

00:02:08: We are talking remote with each other.

00:02:10: I'm sitting in Cologne and you're in Saarbrücken, right?

00:02:13: Absolutely.

00:02:14: So Andreas, what does a clinical bioinformatician actually do?

00:02:19: You see, medical research is damn complex.

00:02:23: We as humans we consist of DNA and then we have many other molecules and these are in cells and we have like roughly thirty trillion cells that our body consists of.

00:02:34: and what most people don't know is we consist of almost the same amount of bacterial cells.

00:02:39: so it's also thirty billion bacterial cells.

00:02:42: and then the human cells.

00:02:43: they form organs they form organ systems and all that is communicating to each other not only human cells to each other but also bacterial with human cells.

00:02:52: And we can measure them by sequencing technologies.

00:02:55: We can take images of organs and we generate tremendous amounts of data.

00:02:59: And whenever it comes to analysis of this data from humans around medical health, then you need a clinical bioinformatician to take care of the data analysis itself.

00:03:10: Nice.

00:03:10: And artificial intelligence plays a major role in your research.

00:03:14: And can the AI you use be compared to... Yes, the AI that we use in our everyday life?

00:03:23: I mean, absolutely.

00:03:24: Our research without AI would not be possible at all.

00:03:27: In research, I would say we are using that already for many, many decades.

00:03:31: This is not what you understand about AI.

00:03:34: It's rather classification technologies such as support vector machines or decision trees that we are using to find patterns in data, for example.

00:03:44: Now the recent successes that you have with these transformer technologies as GBT where you train by huge data sets are not completely comparable.

00:03:53: They are trained on the complete internet and in medical data.

00:03:58: we don't have access to such tremendous data sets at all.

00:04:02: So this means our molecular research data, our images, they are scattered in databases.

00:04:08: and we don't have access to train such large systems at all.

00:04:11: So till we have a system like JetGBT that can really answer some medical research questions highly precisely, this will take time and just one of the reasons is the accessibility of data.

00:04:22: How does AI work?

00:04:24: What does it encompass and how do you use it in your research?

00:04:29: There is no that AI system that does everything.

00:04:33: So you have to distinguish between them.

00:04:35: I just mentioned that you need classification technologies that I use for a long time.

00:04:40: Another class of algorithms that we regularly use is classification technology.

00:04:45: So we do find patterns that a human would never see.

00:04:52: And maybe one comparison that we always have is already a spam filter that we are using.

00:04:57: as kind of an AI system.

00:04:59: But I see one recent trend and this is agendic AI.

00:05:04: So this means we now use more and more these computer language models that we can train like we train research assistants to do certain tasks.

00:05:15: And this is a trend that I see will speed up tremendously over the next years so that we can train computers doing, I would say, easy research tasks in first instance.

00:05:27: In your research, you look at different things.

00:05:30: For example, you investigate the relationship between bacteria and humans, as well as the composition of bacteria that live in and on us, especially on our skin.

00:05:43: What exactly are you looking at and what do you find out there?

00:05:48: I mean microbiota are fascinating.

00:05:50: Microbiota just means the sum of all microorganisms that are in certain specimen types.

00:05:56: This can be in a soil sample, but this can be also in our gut, in our oral cavity or on our skin.

00:06:04: And what we are doing is that we are sequencing all these microbiota.

00:06:08: So per sample it's roughly a five billion basis that we are sequencing and we are doing that for over.

00:06:15: this means that we have like roughly sixty trillion bases that we sequence so far.

00:06:21: And for us it's important to understand which bacteria are in a clinical sample.

00:06:26: Those that are maybe just in healthy samples and not in any patients, maybe they produce something that protects us.

00:06:32: And on the other side also in those patients that have a certain disease, for example, bacteria or viruses that are only in Alzheimer's samples on the skin and the oral cavity and the gut such that you can understand that they produce something which is harmful and this is what makes us a little different.

00:06:50: so we are not so much interested in the bacteria themselves.

00:06:54: We are interested in the biosynthetic gene cluster, so in these parts of the genes that encode for machineries that produce in these natural products.

00:07:03: So we want to understand which metabolites are produced by the bacteria and how we can use them as another therapies, being it anti-infective or in the long term also against neurodegenerative disorders or cancers.

00:07:15: And how does AI help you in this part of your research for new drug candidates especially?

00:07:24: I mentioned already before that tremendous amount of data.

00:07:28: It's a trillion bases that we sequence and it's evidentially clear that neither you nor myself can read those or even make sense out of those.

00:07:36: So there we need the support of computers already.

00:07:40: But then making sense out of this, this is what really matters.

00:07:44: So we are talking about roughly two point five million potential biosynthetic gene clusters that we find in these data.

00:07:52: And we have to understand what they are theoretically producing and what has a therapeutic effect.

00:07:59: And this prioritization task is so important because our colleagues doing the red lab research takes six months, one year, sometimes even longer if it's complicated by a synthetic gene cluster.

00:08:11: to produce these secondary metabolites.

00:08:13: So if we would not tell them precisely which ones they have to test out of these several million candidates that we have, we would have in the end really many disappointed red lab researchers.

00:08:22: And this prioritization task, this is something which is just feasible by using artificial intelligence.

00:08:28: Yeah, I think I can imagine.

00:08:30: If you want to learn more about researching new drug candidates, please listen to episode four of our first season where I talked to Professor Christine Wehmelmann, also from the Hips in Saarbrücken.

00:08:41: But now back to you, Andreas.

00:08:44: Your group isn't just concerned with new active substances.

00:08:48: What else are you researching?

00:08:50: Yeah,

00:08:50: absolutely.

00:08:51: And if you're talking about active substances, it's not only the natural products that I just mentioned before.

00:08:56: So actually I'm interested in RNAs as molecules, most importantly non-coding RNAs.

00:09:01: So how do human genes are regulated?

00:09:04: Maybe even by bacterial RNAs that have an impact on our human genes.

00:09:10: So RNA-based therapeutics is also something which is quite heavy on our research roadmap.

00:09:16: Content-wise, I would say I'm most interested in how we age.

00:09:20: So how our organs age, especially the brain, and to understand the role of the immune system.

00:09:27: And now that we talk on the immune system, it's a gain.

00:09:29: bacteria and viruses that can have an effect on immune cells that have later on consequences as well.

00:09:35: So this is what we want to understand how we age the role of the immune system and what we can do against it.

00:09:42: And by the way, not that we live forever, but that we can make best use of our lifetimes such that our brain for example stays healthy longer and we can enjoy our lifetime much longer as compared if we would have a new degenerative disorder.

00:09:55: Therefore I would recommend our listeners another episode, episode three of our second season.

00:10:01: there I discussed with Professor Martin Korte how infectious diseases can affect our brain.

00:10:07: And on the other hand here you do a basic research but You are also close to patients.

00:10:13: How do these fields work?

00:10:14: How do they differ and what does connect them?

00:10:17: In the end, patient-based research, basic research, they are not mutually exclusive to each other, vice versa.

00:10:24: I think they complement nicely each other.

00:10:26: For example, if we have a basic research question, so we want to find a new drug candidate for Alzheimer's disease, then we need patient samples.

00:10:34: So we have to go to the patient and we have to go to the doctors and ask them, to getting access to these sample types.

00:10:40: Then, of course, we need many molecular technologies.

00:10:43: We have to sequence the bacteria or the human samples.

00:10:47: We have to do data analysis, and we then find eventually, after years of work, drug candidates.

00:10:54: And then it takes, again, a long amount of time until we can play that back, however, to the patient, and then maybe have a candidate that we can test in a phase one or two trial, even so.

00:11:05: basic research, patient-based research, they go hand-in-hand and I think it's a good way it goes.

00:11:11: Can you explain a little more the collaboration with the Saarland University Hospital?

00:11:18: How does it look like?

00:11:19: Absolutely.

00:11:20: I mean, that relation has been grown now for far over a decade and it's supporting each other.

00:11:26: So for example, the clinical doctor generates many data.

00:11:31: images, molecular sequences, and he needs someone that helps in analyzing these data.

00:11:35: And we happily take on the duty and support in this data science task, even if it's on lung cancers, just to support the research there.

00:11:44: On the opposite side, I sometimes need them for one of my studies, maybe patients suffering a certain infection or Alzheimer's patients.

00:11:52: I can go back to the medical doctors and say, hi, hey, I need your support here.

00:11:56: Please get me these samples.

00:11:57: And they happily collect those for me.

00:12:00: And so this relation, this has been drawn and it's a wonderful collaboration that has been established.

00:12:06: And I have to say we formalized that at Saarland a little bit.

00:12:09: So we founded an entity which is called the Pharma Science Hub at the Helmholtz Center for Infectious Diseases at the Saarland University, but also the University Hospital work closely together.

00:12:19: medical researchers natural science researchers and also computational researchers that bring together their competencies.

00:12:26: And especially in these days where we live off a tremendous complex research landscape, it's important that we bring together these different competencies and that they learn speaking the same language.

00:12:38: Yeah, and you learn from each other, I think, yeah.

00:12:40: Absolutely.

00:12:40: I'm thinking as computer science researchers, just in a one and a zero, and for medical researchers, you have many more shades of gray.

00:12:48: Let's return to AI.

00:12:50: Where else can it be beneficial in clinical applications?

00:12:53: And where do you see maybe potential risk or challenges?

00:12:58: First of all, I think there's many opportunities.

00:13:00: So artificial intelligence can make medical researchers tremendously more efficient.

00:13:06: So if you think about having a look at images and finding diagnosis based on images, in many cases, the AI is as good, sometimes even better.

00:13:16: than in clinical doctors.

00:13:17: But we have also to pay attention, I think it's just a supportive function.

00:13:21: So till we have an AI that is even close to replacing a medical doctor, this will take fortunately quite some time.

00:13:28: So there's many risks associated and an AI is only as good as the data it's trained for.

00:13:34: So if you see a patient with rare disease and it has never seen before, then the AI can't get you the right reply.

00:13:42: So all these caveats, we have really to take care on if we're talking about AI in a medical context.

00:13:49: So it's a great tool.

00:13:51: It's a tool that makes medical doctors much more efficient.

00:13:54: It will continue to go.

00:13:56: But till we see a powerful medical AI doctor, this will take some time.

00:14:00: What else do you expect from AI in terms of disease prediction, diagnosis, and maybe even treatment?

00:14:07: I mean by bringing these huge and tremendous data sets together and there's many things that we still can't foresee.

00:14:13: I think by using AI we can make diagnosis of diseases, maybe prevention of diseases much more efficient.

00:14:20: There exist cases where whole trucks have already been designed by artificial or using artificial intelligence.

00:14:28: So the process of how you develop a truck, this is something you can certainly make tremendously more efficient.

00:14:35: by using artificial intelligence as before.

00:14:38: And whether we have a shortcut from like fifteen or seventeen years, it takes currently to bring a drug to the patient to maybe five or ten, this is something which is certainly feasible.

00:14:51: Indeed.

00:14:52: So what is your long-term goal for your research?

00:14:56: The elder you grow... The more I think this changes a little bit.

00:15:00: So first of all, I think we have to do structural things different in research.

00:15:05: So we have to take much more care on training the next generation researchers.

00:15:10: It's clear that we have to...

00:15:11: Not only the AI.

00:15:13: Absolutely.

00:15:14: But it's not only that.

00:15:16: we have to continue, of course, publishing in scientific journals.

00:15:20: We have to understand that science has to become even more open science.

00:15:24: I mentioned that science is complex and it will get even more complex.

00:15:28: So only if we share early time what we know with others, we can do that much more efficiently.

00:15:33: So I think there's many structural things that I want to change or that I think it's necessary to make science much more sustainable.

00:15:42: then it is.

00:15:43: And research-wise, I mean, it's also clear, I want to contribute to understand how the brain ages, the role of bacteria, of viruses, not only an acute infection, but maybe even years or decades after an infection, as we see partially for long COVID, this is something I want to contribute understanding.

00:16:05: only acknowledge that this is a damn complex task.

00:16:08: And if I have a tiny contribution to that, I would be already quite happy and consider myself successful.

00:16:14: Nice.

00:16:14: So, yeah, you do really exciting research.

00:16:17: How do you balance your work?

00:16:19: What do you do in your free time?

00:16:20: And most importantly, of course, my family, I have two little kids and they keep myself and my wife busy all day.

00:16:28: And this is the best thing you can have besides work.

00:16:33: I have to say, in addition, I have kind of an almost addiction, I would say, to doing sports.

00:16:38: So I prefer doing twenty thousand steps over ten thousand steps a day.

00:16:42: I love cycling.

00:16:44: And you say that you have to have a healthy mind and a healthy body.

00:16:47: And I think it's so important and true for whatever disease you can think about.

00:16:52: It's not only obesity.

00:16:53: It's not only diabetes, cardiac diseases, but also new degenerative disorders like Parkinson's or Alzheimer's, where actually sports belongs to the best, if not the best at all.

00:17:05: therapies.

00:17:05: So I think I do even a good thing to my body in that as well.

00:17:09: Finally, what do you like to see from the politicians to ensure that AI can be used safely and sensibly?

00:17:19: And where do you see dangers in these areas?

00:17:22: First of all, I think we should acknowledge that the politics is doing already quite much in the direction of supporting AI research.

00:17:29: So the high-tech agenda of the federal government is just one example, but also at Saarland.

00:17:36: We are doing much in terms of using AI, and now I say already AI in the context of, and this is something which I think it's really important, that we find the right application areas for artificial intelligence.

00:17:49: There's not much that you can think about.

00:17:51: what really matters than your health.

00:17:54: So if we use artificial intelligence together with biotechnology, with pharmaceutical research, then we can really make a tremendous change.

00:18:04: And I think everything which has been done also in our region here, I mean, we are coal mining region originally.

00:18:11: And as you can imagine, there's not much coal mining anymore.

00:18:14: So we have to reinvent ourselves, especially at the Saarland here.

00:18:18: So all these instruments that have been taken care of are important and it's good that they continue.

00:18:23: And besides... I mean, you mentioned also regulation before and regulation is something which is tremendously important.

00:18:28: We could do in biomedical research much more as we currently do because of regulations, because of right regulations.

00:18:36: But we have also to take care on, especially in Germany, that we don't over-regulate.

00:18:41: So I think regulation at the right amount, this is what really matters and what politics should take care

00:18:47: on.

00:18:49: Thank you very much.

00:18:50: Thank you for your time and the exciting talk about your work with AI in clinical

00:18:56: research.

00:18:56: Thank you.

00:18:58: My pleasure, Julian.