Podcast – Using Artificial Intelligence & data to fight dementia

Voice Over:

Welcome to the NIHR Dementia Researcher podcast, brought to you by dementiaresearcher.nihr.ac.uk, in association with Alzheimer’s Research UK and Alzheimer’s Society, supporting early career dementia researchers across the world.

Dr Megan O’Hare:

Hello. Welcome to another podcast brought to you by dementiaresearcher.nihr.ac.uk. I am Megan O’Hare, research associate at UCL, in the Office for the National Director for Dementia Research.

Dr Megan O’Hare:

And I’m pleased to be hosting this podcast today with our special guests, Professor Bart De Strooper, director at the UK DRI, and Professor David Llewellyn from the University of Exeter and the Alan Turing institute.

Dr Megan O’Hare:

And we’re going to be talking today about joining forces, but with a particular focus on the use of artificial intelligence and big data. So, I’m going to say probably most of you have heard of the UK Dementia Research Institute, or the UK DRI, but here’s a quick, succinct summary that I basically just lifted from their website.

Dr Megan O’Hare:

Launched in 2017, the UK DRI is the single biggest investment the UK has ever made in dementia. Thanks to 290 million pounds from the three founding funders the Medical Research Council, the MRC, Alzheimer’s Society, and Alzheimer’s Research UK.

Dr Megan O’Hare:

The UK DRI breaks new ground by bringing together world leading expertise in biomedical care and translational dementia research. In a national institute, currently made up of over 600 researchers, plus 150 students, and a support team of over 50 people. And I think it’s six or seven sites. I think six sites, and then-

Professor Bart De Strooper:

It’s six sites, and seven centres.

Dr Megan O’Hare:

Yes, that’s it.

Professor Bart De Strooper:

So [crosstalk 00:01:59] centres, a care and tech, and a basic research centre.

Dr Megan O’Hare:

Yeah. And so, as we can see from that or hear from that, the UK DRI is a massive organization, but it also part of the wider UK dementia research landscape and knows the value of good collaboration. And that’s basically part of what we’re going to talk about today is collaboration and partnerships and how important they are, especially in dementia research.

Dr Megan O’Hare:

So, the ambitious new partnership that we’re going to be talking about today is between the UK DRI and the Deep Dementia Phenotyping Network, so the DEMON network, and that’s why we’ve got David on today.

Dr Megan O’Hare:

This partnership aims to harness data science and artificial intelligence to full advantage to rapidly speed up the transformation of data into clinical and biologically relevant knowledge in neurodegeneration research. So, I am delighted to introduce our two guests. Hello.

Professor David Llewellyn:

Hi, Megan.

Professor Bart De Strooper:

Hi, Megan.

Dr Megan O’Hare:

Hi, hi. They are here, really, with me. So maybe we can start with you, Bart, by asking you to introduce yourself, a bit about your background, and how you ended up being the director of the UK DRI?

Professor Bart De Strooper:

Yeah. Well, that could be a very long story, but I’m an Alzheimer’s research for a long time, and I was working in Belgium, in Leuven, created here a centre focusing on neuroscience and Alzheimer’s.

Professor Bart De Strooper:

But then, in 2016, I got the news that UK wanted to launch this fantastic initiative. And it’s a long dream of me to get things working together and to get researchers working together.

Professor Bart De Strooper:

And I’m also already dreaming for, from the very beginning, to get the same attention for dementia as for cancer. So, for cancer, we have all these fantastic research centres. The most exciting new, fundamental research is immediately applied. And people don’t realize, but there are close to 50 million people with dementia in the world, and there are close to 50 million … It’s even less, 40 million with cancer.

Professor Bart De Strooper:

So, the problem, medically, is equally big. And so that’s why I thought, when UK did this call, “This is something I want to do in the last part of my career.” And so, I applied and … Yeah, well, the rest is water under the bridge, I would say, or is a long story.

Professor Bart De Strooper:

But I ended with being offered that position, and I felt very honoured, and so I moved into it. And now, three years later, it’s amazing. It’s an amazing institute. We have now 60, about 60 group leaders, very young ones. We attracted them from all over the world.

Professor Bart De Strooper:

We attracted them from different disciplines, and so the machine has started to run. And COVID has been a big problem, but even under COVID times, we are still progressing. Now we are looking around for collaborations because that is one of the main messages we got.

Professor Bart De Strooper:

Of course, an institute of 600 is big, but you cannot do all the research. And so, we are now that institute, and now we are reaching out and trying to find all the other fantastic research happening in the UK, and to link ourselves to them. And so, data science and artificial intelligence is obviously a big, big need.

Dr Megan O’Hare:

Yeah. I think, when you said “working together,” I think that’s why the UK DRI is just such a fantastic place, because like I said at the beginning, it brings in biomedical care and translational research, and that’s sort of how you dreamed the Dementia Research Institute would be, bringing all those bits together.

Dr Megan O’Hare:

You know, “What if we could do this?” And it is doing that, and it is working really well. When I went to write the little blurb about it, I took it from the website, and it said 350 researchers. Then it got picked up and then it’s already grown to 600, so …

Professor Bart De Strooper:

Yeah. I told our web person that he had to update it.

Dr Megan O’Hare:

Yeah. Yeah, he does.

Professor Bart De Strooper:

But that is the nice thing. We can start the institute. It changes every three months, everything changes. It’s an exciting time. The COVID is really, really … What do you call it? Cold water on the whole machine, but it is so strong that we are still working, and we will grow. We will get there.

Professor Bart De Strooper:

And so, the other thing is it’s a multi-purpose institute, so you have to see it at the building, but every centre, every layer is placed in another university. So, it’s really multi-purpose, because that gives us also outreach to what happens in those universities. So that’s also an example of the collaborative aims of the institute.

Professor Bart De Strooper:

Now, with all this knowledge in the one research, we keep it where it was, and we strengthen it, and we try to bring these centres working with each other together.

Dr Megan O’Hare:

Yeah, and that’s sort of where David, you come in, although you’re not part of the UK DRI. You are … What would you call it? A network that’s also national, and you’re being brought in as another bit of-

Professor David Llewellyn:

Piece of the jigsaw puzzle.

Dr Megan O’Hare:

Yeah, piece of the jigsaw puzzle.

Professor David Llewellyn:

Yeah. I think you’re right, in the sense that it’s all about collaboration and innovation, Megan. We did set it up, initially, as a national network, but we had so many inquiries from people across the globe, that the DEMON network’s now truly an international network.

Professor David Llewellyn:

I mean, we started it because of our frustration about the way that things were happening in silos. And some of what you said, Bart, it kind of resonates with what we were thinking a couple of years ago as well.

Professor David Llewellyn:

And we’re looking at the research landscape in the UK, but also globally, and what the gaps are, and what we thought the opportunities were. And I got a fellowship at the Alan Turing Institute, where obviously the focus is a lot on innovation methodologically, and how to use data and the potential of AI and machine learning.

Professor David Llewellyn:

And I was talking to Carol Routledge, who was director of Alzheimer’s Research UK at the time. We started to see a change in the kinds of applications that were being submitted, so people were seeing that they were going to use big data, or they were going to use different approaches.

Professor David Llewellyn:

And they were interesting, but they normally didn’t have the right kind of teams. They were talented biomedical researchers that weren’t really talking to the computer scientists or the data scientists.

Professor David Llewellyn:

And often, there weren’t clinicians involved in the projects as well. So sometimes, we were a bit concerned with the clinical relevance with what was being done. So, from a selfish perspective, we decided that, in our work, and in the way we wanted to work with other people, that we wanted something different. We wanted a different way of working.

Professor David Llewellyn:

And we were lucky in that we got seed funding from the Alan Turing Institute, first, to start something. And we’ve always envisaged it being around an infrastructure for people, so other major initiatives … Obviously, we have the DRI, but we have other things as well, like Dementias Platform UK, which provides complementary infrastructure, making data more easily available.

Professor David Llewellyn:

And we thought, “Wouldn’t it be amazing if we set up something which was a kind of platform for the talent and the innovators to come together?” And the idea, initially, wasn’t really very well-formed. I mean, as I say, it started off as, we would build something that would become a national network, but it’s kind of evolved.

Professor David Llewellyn:

We only launched in November, and just yesterday, we had our 600th member join the network. So, it’s amazing how quickly it’s grown, and we’re recruiting at lots of different positions to help drive that agenda forward.

Professor David Llewellyn:

We’ve got things like practical working groups. But we were delighted to officially partner with the DRI, because we have got a global network of innovators who are fascinated by doing things differently.

Professor David Llewellyn:

We want to achieve things like disease-modifying treatments in dementia. I mean, you can mope about and be quite depressed if you’re a dementia researcher. We’ve continued to fail, and we just have an appetite to do things differently. That’s why we’re so excited to work with the DRI, and to work with others as well. But it’s all about getting out of those traditional silos and doing things differently.

Dr Megan O’Hare:

So, you said it started with, you were getting applications from biochemists, biologists, with great ideas, but didn’t have the right people behind them, or know the right people sort of thing. So, is it sort of that the DEMON network is kind of matchmaking with big data scientists, or can anyone join? Or is that you have a pool of big data scientists that can be …

Professor David Llewellyn:

Yeah, no. No, we deliberately … I mean, it’s designed to encourage collaboration and new ideas, dangerous ideas. So, we want people to join who don’t know anything about dementia. Maybe they know all about transfer learning, or other forms of machine learning, or artificial intelligence.

Professor David Llewellyn:

Maybe they’ve got novel methods and actually want to collaborate with other people who know all about dementia, or … There are just different ways of doing things, and I think without genuine diversity, we really struggle to get new ways of working in place.

Professor David Llewellyn:

So no, it’s completely open. And we have a public patient involvement group. We have a clinical advisory group. There’s lots of resources that we can develop, where we have early career research leads. The ambition is to develop resources, learning materials, working groups. We’ll be launching our monthly lecture series very shortly, and we’re doing that in partnership with the DRI, so that every other seminar series is going to be directly relevant to the DRI.

Professor David Llewellyn:

So, we’re thinking about practical ways to get people working together differently, and it may be that you know all about computational biology, but you don’t know a lot about maybe the patient pathway for dementia patients. So, people will go off in different directions, and they won’t be interested in something, and that’s okay. We just want to mix things up and see what happens, really.

Dr Megan O’Hare:

Yeah. You mentioned a couple of times about having clinicians onboard and how important that is, and I don’t actually think, at the UK DRI, there are … There’s not a clinical section, is there?

Professor Bart De Strooper:

There is. There is, I think …

Dr Megan O’Hare:

There is? Okay.

Professor Bart De Strooper:

I think half of our directors are active clinicians, and we have a lot of clinical people who have a position in the clinic, and do research in the DRI. So, it’s also a very broad institute. We have a focus on disease, but we approach it from all corners.

Professor Bart De Strooper:

So, what we miss now is basically what DEMON is going to give us. That’s a very strong data science unit. We are building that, so we are going to recruit a data science director and we are building a whole team around, and then we have also a couple of group leaders who have started a bioinformatics data science team throughout the DRI, where we will standardize our data collection, where we make it people to access each other’s data.

Professor Bart De Strooper:

There’s a lot of groundwork which has to be done, to make that all effective, and we do that in collaboration with the DEMON, of course. So, yeah. I mean, we have a lot of clinical data in the DRI. We have also a care and tech centre, which is all about patients.

Professor Bart De Strooper:

So that researchers, we go … We have even our own houses, and we have then people with dementia coming in, and then we follow those patients over weeks or months even, with all kind of tools, modern recording tools and to see how we can help these patients with their life at home, et cetera.

Professor Bart De Strooper:

So, one of the things I find fascinating, [inaudible 00:15:55], so you know that with these patients which have dementia, one of the most frequent problems which requires hospitalization is a bladder infection, because these patients are diagnosed much too late. They’re extremely sensitive for that.

Professor Bart De Strooper:

And so, if you can record that very early before they become sick, you can avoid hospitalization and a lot of confusion. And so, one of the projects is to have a special stick in the toilet linked to an alarm system, so that once there are bugs in the urine, there can be immediately a warning signal, and we can have the nurses or someone [inaudible 00:16:40] at home can immediately start the treatment.

Professor Bart De Strooper:

So, these simple things are also part of the DRI. So, people always think that we do these extremely … and we do that. These extremely ground-breaking, cutting edge research, in molecular biology and gene therapy and these types of things, but also this type of research is very important.

Professor Bart De Strooper:

And so, the real challenge is to link all this information, because it’s all part of the same picture. And so, I am a molecular biologist, so for me, it’s not obvious to see how that has to be done.

Professor Bart De Strooper:

And I think that people who are real good in data science don’t think about urinary infection in old people, or have no clue how that could be influenced by genetics of these patients, or by the disease process itself.

Professor Bart De Strooper:

So that’s things which we understand, but we don’t understand how we can bring all this data we are generating in molecular biology and in the clinics, how to integrate that in a coherent model for Alzheimer’s disease or another dementia form. And so that’s, for me, the big challenge to do, and that’s where we look for partners.

Dr Megan O’Hare:

Yeah. And I think, at the beginning, one of the key things was transformation of data into clinical and biologically relevant knowledge. And that’s sort of the bit, isn’t it? The joining together of those things. Yeah.

Professor Bart De Strooper:

It goes in two directions. That’s really also, I liked to hear from David that some of these people busy with data, they generate models. And sometimes, my feeling is, in many cases, that it comes out of the blue. They come up with this type of model based on all the data.

Professor Bart De Strooper:

But what I’m saying [inaudible 00:18:30] about, and even if the data scientists say, “A lot of data, you will always find something.” I agree, but I’m still in that model, “Crap in, crap out.” So if you really understand very well where the data are coming in, if it’s quality checks there, if you go a little bit with us to understand what we are doing, then you as a data scientist will make much, much, so much better models.

Professor Bart De Strooper:

And it’s a reiteration. So, you talk while you’re analysing those data, “What does that mean? I don’t get it. Is this the way I should understand this?” You get this kind of conversation while you are doing your data science. And we, as biologists, get this conversation. While you are developing that model, we can give input and we can understand the real reason. And that’s where new knowledge will come.

Dr Megan O’Hare:

When I discovered I’d be hosting this podcast, I thought to myself, “I should probably actually work out what artificial intelligence is.” Obviously, we use the word a lot, and you think of that Will Smith film.

Dr Megan O’Hare:

So, I had a look around, and there’s the Turing test, which is basically the test of a machine’s ability to exhibit intelligent behaviour equivalent to or indistinguishable from that of a human.

Dr Megan O’Hare:

But I wondered, because you’re both coming at it from different places, how you view AI, and how you think it will impact dementia research? Maybe David, we’ll start with you?

Professor David Llewellyn:

Well, it’s a difficult thing to pin down, because obviously, it means different things to different people. But I think in its broader sense, artificial intelligence is the idea that we’ll create machines that think for themselves, and are able to exhibit intelligent characteristics, without us pulling levers and manually inputting, and prodding and poking them into what they should be doing.

Professor David Llewellyn:

And the most ambitious of artificial intelligence is the creation of a new, digital, sentient being, which will take over the world and enslave us, and that’s the thing that some people worry about. The point at which machines don’t just match our performance but start to exceed our performance.

Professor David Llewellyn:

So, that is a fascinating and yet unachieved ambition, that not everyone thinks we should be aiming for either. I don’t think many people are arguing that we’ve created a super intelligence yet, but it’s possible. It’s possible.

Professor David Llewellyn:

But I think, to many people, artificial intelligence is a more down-to-earth … It’s something about, I think, machine learning really, where we’ve trained machines up to do narrower tasks very well. And we’ve already seen computers outperforming humans with many things.

Professor David Llewellyn:

I mean, there are famous examples of thinking computers performing really well at chess, or AlphaGo and those kinds of games. And it’s easiest to do, obviously, with things which have very defined rules, and that’s where you’ll get the progress most quickly.

Professor David Llewellyn:

How does that translate to something a lot messier and more complicated, and less cerebral, like dementia itself? We have an aging population, messy and overlapping pathologies that are themselves not completely understood.

Professor David Llewellyn:

And then we try and model people’s lifespans and understand how these cognitive deficits start to accumulate, often in a delayed way in response to the accumulation of pathology. And then layer on top the additional complexity that, even if we understood what was going on completely, which I think it’s fair to say we don’t, then it doesn’t tell us automatically how to intervene.

Professor David Llewellyn:

So, we think about how to apply these kinds of intelligent approaches to things like drug discovery and clinical trials. And so, the application of machine learning is a very practical way of applying a limited form of artificial intelligence.

Professor David Llewellyn:

I think that’s my take on it, though, that artificial intelligence means … It’s this kind of spectrum of relating to how ambitious what you’re trying to achieve with it is.

Dr Megan O’Hare:

I think the classic ones are sort of using it to analyse brain scans and use it over time to track people’s progress, ultimately also to start to predict earlier on in the disease pathway, certain things. And that, obviously, comes with machine learning. So far-

Professor David Llewellyn:

I mean, you’re right. Imaging is being an early success story for machine learning, but it’s often being used in other areas first. I mean, Bart was quite right in saying that looking to other disciplines and aiming for us to match or exceed what they’re able to do in cancer and so on.

Professor David Llewellyn:

And things like deep neural networks have been used to analyse images and detect skin cancer with comparable or even superior performance than dermatologists. I mean, those kinds of examples are … You can build narrow intelligences, which are essentially data-driven models, but they don’t tell you … They’re very superficial models in one sense.

Professor David Llewellyn:

They’re very useful, but they don’t tell you why things are happening, typically. And in terms of biological mechanisms, and I think that resonated with what you said before, Bart, about going beyond just a purely data-driven approach.

Professor David Llewellyn:

I don’t think we’re very good at developing causal models, because I think what we do is, we detect patterns in the data. But often, the data aren’t sufficient to allow us to unpack causal questions.

Professor David Llewellyn:

Some neat things we can do by using, say, genetics and genetic variation, as essentially a form of natural, randomized control trial for exposures and so on. So, there’s so much more that we’ve yet to even think of how to do things differently. I mean, I think we’re taking baby steps at the moment, in terms of what’s possible.

Dr Megan O’Hare:

And I guess that’s where the collaboration is key, isn’t it? That you maybe know how to utilize AI, but wouldn’t … or looking for the right question, and that’s where the collaborations come in, is that Bart’s centre will come up with the questions?

Professor Bart De Strooper:

Yeah. Well, yeah, but it’s an interactive way, these things. That’s always the problem with complex … If you’re in unknown territory, there is not one guide who knows exactly the way, because it has never been charted.

Professor Bart De Strooper:

So, I see it more as an interactive process where we have some idea, say, “Let’s go in that direction because we believe, basically, that that’s a good direction.” But while you are walking, you are probably going to another direction, because you suddenly start to see how the landscape evolves. So that’s one take.

Professor Bart De Strooper:

The second take is that I think it is high time that artificial intelligence, data science … I don’t know how it’s in the cancer field, but in the dementia field, it’s very frustrating.

Professor Bart De Strooper:

So, I think that data science becomes very rapidly easier, and it’s same with genetics, to be honest. It becomes very descriptive. So, what’s very good in data science, and that’s what I hope that will happen, is to recognize patterns in data which are complicated, that we are not able to see that with normal human brain or human eye, but where these patterns are still clearly present.

Professor Bart De Strooper:

So that’s one thing which I love, and I will say is artificial intelligence, because it uses that and certain learning from people, and that is fed through the machine, and the machine is then able to find algorithms which mimic that learning, and then able to apply it to thousands of data points, while your brain is only able to do that to 50 data points, and then you get much clearer and much crisper insights into what’s happened.

Professor Bart De Strooper:

That’s what biologists do the whole time, when they generate … Biologists are fascinated by data. They do experiments and they have all this data and they’re very happy. And then when they want to make a paper, they try to find the pattern in those data, and that’s what they call “insight.”

Professor Bart De Strooper:

And so there, I think artificial intelligence is going to help us a lot. But it’s also the limitation, because even when you see a pattern and you describe it, you’re still not really understanding what’s happening there.

Professor Bart De Strooper:

And that’s where my training as a molecular biologist becomes very strong. We are very, very much focused on mechanisms. So, for us, the bigger thing is, you understand how a protein has a structure and how it interacts with another protein, and how then that activates how a muscle move, et cetera.

Professor Bart De Strooper:

So that’s, for us, real insight. And once we have that insight, it becomes rather easy to interfere, because we see how the model is working, and now we can see if we change that little thing, maybe we can cure a disease or whatever.

Professor Bart De Strooper:

So, I think that the data scientists are happy when they have that pattern and say, “Look, according to that pattern, if you push here, the pattern will move in that direction.” While the molecular neurobiologist wants to know why, if you push a little bit here, you make a mutation here, why exactly, mechanistically, that changes?

Professor Bart De Strooper:

And so that’s why these two fields are so different. It’s like morphologists, pathologists, and molecular neurobiologists too. Pathologists are very happy with a disease that they can see in their microscope.

Professor Bart De Strooper:

But for us molecular biologists, yeah. I mean, it doesn’t mean nothing if there is [inaudible 00:29:09] there. What’s the molecules there? What’s happening there? So, I think that combining these two approaches, this more descriptive, pattern recognition approach, allows you to get much faster and deeper into complex mechanisms.

Professor Bart De Strooper:

So, the disease advances of molecular biology and how it’s done at the moment is, it’s very simplistic. We use a … I’m now, of course, simplifying it a little bit. I apologize to my colleagues. But we take a cell. We put DNA in it. We express the protein, and now we see things happening, of course, and then we think that’s biology.

Professor Bart De Strooper:

Of course, biology is much more complicated. So, the next step in molecular biology was … If I’m boring you, please interrupt me. The next step was the model stuff. When we changed the gene and then we saw things changing in the model, we said, “We understand the gene.”

Professor Bart De Strooper:

But of course, then we see that doesn’t translate into dementia, because if you think about dementia, that’s a very complicated process. First of all, dementia is a symptom. It’s very bad to call dementia a disease, but that’s another discussion.

Professor Bart De Strooper:

But then, you start to think about how the human brain works. First of all, there is no good model for a human brain. Second, you have all these genetic background [inaudible 00:30:29] which you can’t mimic in a model. You can only mimic that in a human, the genetic diversity of a population.

Professor Bart De Strooper:

And so then, we start to understand this, and that’s why molecular biology is developing a single cell analysis, these more complex, organized models, blah, blah, blah. But that faces us now with the problem, you get all this data now, and we’re of course, very excited, but we don’t understand it anymore, because we don’t have this cause-consequence relationship anymore.

Professor Bart De Strooper:

And so, the pattern specialists will be happy, or will be very helpful to say, “Look. You look through this data in this way. There is a correlation between those things.” And that’s the big thing, correlation, that you have to remember. That’s the problem with descriptive science. That’s what you have when you do MRIs, when you do clinical work, the best thing you can do is correlative, correlation.

Professor Bart De Strooper:

But once you have defined that, you create a hypothesis. And that hypothesis can be back-tested by molecular biologists, because then we can say, “Okay, if this correlation holds true, and we take out this or this piece of the correlative network, we would predict that this or this happens.”

Professor Bart De Strooper:

And if we can prove that this happens, then we know that the correlation is probably a causal relationship, and we are home. Or we can tell the data scientists, “Look, guys. It’s very nice. You make here a causal relationship, or you suggest a correlation or …” And many suggest casual relationships. “You suggest that, but if we take it out, we just get the reverse effect.”

Professor Bart De Strooper:

And that happens. And so that’s interesting, because the data scientists never hear about that 10 years later, because we read different journals. But in this case, in the DEMON network, you will have predicted something to this biologist who is going to take it to the test. And of course, you as a data scientist want to know what happens, whether your prediction holds true.

Professor Bart De Strooper:

And the biologist will come back and say, “No, this is not holding true.” And the data scientist will go back to his machine and artificial intelligence and everything to see whether maybe that pattern is a little bit different. That’s what’s going to happen.

Professor David Llewellyn:

I mean, the key principle there is testability, isn’t it?

Professor Bart De Strooper:

Yes.

Professor David Llewellyn:

So whatever type of model or prediction you’re making, if it’s not testable, then it’s not going to fuel progress, ultimately.

Professor Bart De Strooper:

Yeah, exactly.

Professor David Llewellyn:

I mean, I love that. I love that idea of going backwards and forwards and testing, challenging each other. In a nice, constructive way, but nonetheless, a challenge. But I don’t think we should fall into the trap of assuming that all data science is going to be just descriptive, you know what I mean? But a lot of it has been. But I think-

Professor Bart De Strooper:

But I see an evolution there. So of course, you can make a model then and say it is predictive. And so, certainly for the dementia field, we don’t have much predictive models. So, I think this new confirmation and this testing of your models before you become really … What do you call it? Functional.

Professor Bart De Strooper:

Because I think at a certain moment, that’s of course the goal of these type of things, that you get an in-silico model of the disease. And that will be … It’s probably the only way to get all the data integrated, because in our animal models, we are always limited for the reasons I’m saying.

Professor Bart De Strooper:

But so, if we give you sufficient tests of the different steps in your modelling, at the end, you will be predictive and you will be able to say, “Look, if you change this with the drug, you will get that outcome.” That’s the ultimate goal, and that’s why we want to collaborate, because it will be more effective than anything we are doing at the moment.

Professor David Llewellyn:

Well, that’s the acid test ultimately, isn’t it? If you can understand what’s going on and if you can predict the outcomes of future clinical trials, then you’re in an unbelievable position to develop disease-modifying treatment.

Professor Bart De Strooper:

Everything. It changes everything. It’s not only upstream towards therapy, but also to our type of research. There will be a moment, and that’s also the vision of the DRI, there is a moment that most of the research will be in silico.

Professor Bart De Strooper:

So, once we have really good models for all these mechanisms, these biological mechanisms, once we have mathematical descriptions of physiological processes, you will be much more biologists using all these models to test your hypothesis, and then you will do predictions based on these models.

Professor Bart De Strooper:

And then, based on how high the confidence is in these models, you will either go immediately to human testing, or maybe you need still this intermediary step, but I will … I think that in 10 years, I foresee for me now, still going on, that a lot of the classical biology will be replaced by this combination of in silico prediction and then testing. In 20 years, it’s probably going to be … The classical biologists will be a rare species.

Professor David Llewellyn:

Well, if you listen to some people, artificial intelligence will make us all redundant.

Professor Bart De Strooper:

Yeah, yeah.

Dr Megan O’Hare:

You mentioned about drug testing and clinical trials. I just wondered what sort of stage we were at with using big data and AI for that, because you were saying maybe you can predict the outcomes of a clinical trial. But at the moment, are you able to use big data to say things like, “This population might benefit from it more based on these predictions or this risk factor”? Are you able to do that right now?

Professor David Llewellyn:

Yeah. I think we’re starting to see earlier example of data science-informed trials recruitment, targeting more specific populations. I mean, one of the things that really challenged trials in the dementia field has been the incredible cost. So, having to sift through dozens or even 100 people to recruit one suitable person for a trial in a targeted fashion, that’s been incredibly wasteful.

Professor David Llewellyn:

And then, we have a lot of people in trials who don’t get worse during a six-month or three-year trials period. So, you’re trying to make them better, but actually, they weren’t really the right kind of patient again. A lot of our trials effects that we’re trying to discover are kind of watered down by the nature of the samples that were recruited.

Professor David Llewellyn:

So, I mean, we’ve started to develop different recruitment tools, using a data-driven approach, to try and help with that problem, and I think a lot more could be done there. The way in which trials are conducted, though, could be very different.

Professor David Llewellyn:

And there is interest. It’s been a bit slow to develop, but there is interest in more targeted treatments and adaptive trials designs. I think that could be transformative. But it’s not a trivial thing. Think about how tightly regulated trials are, and how nervous it’s going to make the ethics panels, the idea that those treatments are going to be decided upon, on the fly, by an artificial intelligence or form of.

Professor David Llewellyn:

I mean, it’s not going to be easy. But the potential payoff is, of course, enormous. I mean, clinicians, when they’re dealing with patients, they will tailor what they do depending on what the patient needs. And why wouldn’t we do trials in that kind of way? So, yeah. It’s going to take a while. I think progress in that area has been quite frustratingly slow, I feel.

Dr Megan O’Hare:

Do you find a lot of resistance to AI? You said that the ethics committees might balk at the idea that it would be a machine deciding something, but I mean, you’ve said it’s just a pattern recognition which humans do, and they’re not infallible.

Professor David Llewellyn:

Well, there should be a lot of resistance to AI, because if you don’t understand what it is, then from an ethical perspective, it should raise concerns. And let’s face it. A lot of the models that have been built have bias incorporated in them. And that could have real world consequences.

Dr Megan O’Hare:

Bias because they were built by a human. Or because …

Professor David Llewellyn:

No, because of the nature of the … It’s normally, I feel, that the bias incorporated in the models is a reflection of the bias incorporated in the data that the models were trained on. So, if we use a nice, easily obtained sample of 10,000 volunteers, then yeah, it’s big data, but is that going to work in the general population? And the answer is, it will probably serve best the type of people who volunteer for those kinds of research studies.

Professor Bart De Strooper:

Definitely.

Dr Megan O’Hare:

That’s always the problem.

Professor David Llewellyn:

And we know that we’re hearing constantly about technologies which potentially certain groups. We’re all concerned about … I mean, we’re concerned about unconscious bias more generally in the workplace, but I think we crystallize biases, potentially, through the development of models, unless we’re very careful about what we’re doing.

Professor David Llewellyn:

It does go back to that idea of a deeper understanding of the data. I think it’s a slight variant on what you were saying, Bart, in terms of mechanistic understanding. I think there’s also that kind of broader, human understanding of the limitations of data and the limitations of models.

Professor Bart De Strooper:

I am also thinking about the … For me, the biggest limitation is dementia as a disease. So, most of the patients we incorporate in studies, even genetic studies, are diagnosed based on cognitive failure. But so, we know that Alzheimer’s starts 10-15 years before you have those measurements failing. So, there are probably other things, which we still don’t know, which are failing very early on, but nobody’s measuring that.

Professor Bart De Strooper:

So you get the population, where in the total population, there are probably a lot of people of a certain age, which have already Alzheimer’s disease, but don’t show the clinical [inaudible 00:41:46] part, so you get all this enrichment of patients in your consult group, and then reserve in your test group, with people who have cognitive problems, but if you are studying Alzheimer’s disease, which is a widely defined molecular pathology, but you take dementia as the paradigm, then you will include a lot of people who have dementia, but because of another disease.

Professor Bart De Strooper:

Alpha-synuclein, [inaudible 00:42:19] pathology. I mean, there are many, many causes of dementia, and we know that. And so that makes big papers and clinical trials, because you treat the patient with a drug which is not really aiming at the molecular biology that’s tackled by the drug. And so-

Dr Megan O’Hare:

This is where AI and machine learning can come in again, to predict earlier, or pick up people earlier.

Professor Bart De Strooper:

Yeah, but I want to just make the point, so if you are in machine learning and data science, you just use the same data as these clinicians or this pharmacological industry is using, when they are talking about dementia.

Professor Bart De Strooper:

And you will do artificial intelligence and data science on dementia, and not considering all this mechanistically related insight. And so, that’s also why I think that we need to focus much more.

Professor Bart De Strooper:

I honestly think that data science and artificial intelligence is a very, very, very early phase in [inaudible 00:43:27]. I want to … The reason is double. First, they can relatively make [inaudible 00:43:34] in dementia. And second, dementia is a very immature field.

Professor Bart De Strooper:

The knowledge in that field is so much behind anything what we could know if we would have invested more. So that’s also one of my favourites. Think about publications in PubMed, which for me, the database, the database of science.

Professor Bart De Strooper:

So, PubMed contains all the publications which are published in journals like Nature, Science, but also all the smaller journals. And so, we can count the number of papers published with Alzheimer’s or with dementia over the last 100 years, and we can do the same for cancer.

Professor Bart De Strooper:

And so, for dementia, it’s close to 200,000 papers. And for cancer, it’s 3.5 million. So, there is a database which is 15 times bigger for cancer than for Alzheimer’s disease. So, if you tried to use all this data science from cancer into dementia research, yeah. Well, you have 50 times less good quality, because there’s also, good data are characterized by redundancy.

Professor Bart De Strooper:

So, if you have the data confirmed to two or three independent studies, or a molecular link confirmed to two or three independent studies, that’s good data. If you have one single study, for the biases that David just mentioned, to one single study, then it holds true within that small subgroup where you did your experiments.

Professor Bart De Strooper:

So, in cancer, most of the molecular links are confirmed two, three, four, five times. In dementia, most of the molecular links are confirmed only once. So, the data scientist works with those data which are much weaker than the type of data you can get if you work in the cancer.

Professor Bart De Strooper:

That changes a lot of the people in the data world to the cancer field. There’s more room to do strange things, and there’s much better data. And so that’s one of the missions with have with the DRI, to make good quality data available to good data scientists and convince them that it’s worthwhile to work with us.

Dr Megan O’Hare:

In a way, that’s a good thing, isn’t it, that they’ve come in earlier?

Professor Bart De Strooper:

Yes, absolutely.

Dr Megan O’Hare:

So, they can influence what data you’re collecting and influence the clinical pathways so they can influence how your diagnosis, so then you’ve got an early population. And perhaps cancer would have benefited if it had all that, 50 years earlier.

Professor Bart De Strooper:

So, by doing these initiatives, we will catch up with … That’s the [inaudible 00:46:15] of, if you lag behind, it’s much easier to catch up than the front line. So, we are now catching up. And that will be a revolution in dementia.

Professor David Llewellyn:

I mean, I think it’ll be better, but only for certain types of people, this environment with dementia research, and trying to work differently. I think if you want to join something far more developed and do something very specific, and target incremental gains, then I think a field like cancer would be better for you.

Professor David Llewellyn:

But I think the people who want to crash about and try things, there’s so … I mean, it’s not a blank canvas. That would be overstating it, but there’s so much room for innovation. I think that’ll be really exciting to the next generation of researcher.

Professor Bart De Strooper:

Yeah.

Dr Megan O’Hare:

I want to get back in the lab now.

Professor Bart De Strooper:

Yeah, of course. It’s a great time. Really, I am saying that to all the young people. This is the time. It’s the time of brain scientists, because that’s also another interesting spin to this whole discussion.

Professor Bart De Strooper:

So, artificial intelligence and data science, but artificial intelligence for sure, I mean, the big example is the brain. And so, I mean, if you think about the real exciting research which is going to be characterizing this century, it’s the brain. It’s brain research.

Professor Bart De Strooper:

We will understand, at the end of this century, the brain like we understood, at the end of the previous century, the molecular biology, the genetics, the chromosomes, all these things. This is what will be in the history books in 2000 years. They will say, “That was the century when the human brain was understood, how it works.”

Professor Bart De Strooper:

And so, people say dementia is a bit depressing for several reasons. When I started, it was a bit depressing. There was this taboo around dementia and it’s really, “It’s descriptive and it’s old age and you cannot do anything anymore.”

Professor Bart De Strooper:

I don’t agree anymore with that type of thinking. This is a very fascinating area of research, and it’s an unknown territory as David says. So, for the real pioneers under us, this is the place to be.

Professor Bart De Strooper:

And so, if you are interested in the brain and you are interested in artificial intelligence, and you want also to do something meaningful for society, apart from doing this basic research, which is absolutely important, but if you want to do it somewhere else and apply it, go, then dementia research is really the place to be.

Dr Megan O’Hare:

Wow.

Professor Bart De Strooper:

And I mean it.

Dr Megan O’Hare:

I believe you.

Professor David Llewellyn:

Yeah, I believe you too, Bart.

Dr Megan O’Hare:

So, we’re kind of coming to the end now. Are there any other comments you’d like to make about collaboration, AI? We haven’t really touched that much, actually, on the use of AI in the care sector, which you did kind of mention earlier about UTIs and testing. Is there anything you’d like to say about that?

Professor Bart De Strooper:

They’re working on that.

Dr Megan O’Hare:

Yeah?

Professor Bart De Strooper:

So, the clinic is already working on that. So, they do a lot of data science. They have a couple of engineers there, and so they are thinking about algorithms basically, so you have this feedback group that says … So that’s really already quite well going, and I think we will get, very rapidly also, interactions with the DEMON network there.

Professor Bart De Strooper:

But I think the challenge for me is to get sense of all this other data we are generating. It’s such a rich source, the DRI, so we have the sleep research. Sleep is another fascinating area of brain and dementia activity.

Professor Bart De Strooper:

So, we have both basic research and models, but also, we have a whole setup on sleep in humans, in the sleep clinic, which we’ve got a lot of data. So, then we have a big brain atlas project, where we are going to analyse all the molecular changes over the whole brain, in a couple of brains of patients, and so we have all this data.

Professor Bart De Strooper:

So, everything, yeah, well, trying to help us. So that is the big question. We have this mouse modelling where we try to mimic certain aspects of the process. We have the human data, so how much do they overlap? What’s different? What’s similar? That’s the real hard … Well, I need to be careful. That’s really a hard question to crack.

Professor Bart De Strooper:

And so, once that’s done, we know what’s predictive in a mouse model, so then we can help the data scientists when they make their model in the human data, then we say in the mouse, we can mimic that part. So, it’s really an amazing stuff.

Professor Bart De Strooper:

We are also starting to do gene therapy in patients. That will give us a lot of data. So, if this data, if we find a failure, we can have the data scientist come into our data and helping to explain, why did it fail? Why did it work in patient A and not in B? That’s a failed drug trial, because 50% is absolutely [inaudible 00:51:42]

Professor Bart De Strooper:

But help us to understand, what’s different in-patient A versus patient B? And so, there’s so many questions. Yeah, so it’s a pity we couldn’t discuss about all these things, but yeah. Time is limited, of course.

Dr Megan O’Hare:

We can have you on again.

Professor Bart De Strooper:

Yeah. Well, if you want.

Dr Megan O’Hare:

David, any final comments for you?

Professor David Llewellyn:

Yeah. I mean, it’s wonderful to hear about the exciting things in the pipeline, and the opportunities that there are, within the DRI but also in partnership with the DRI. I agree with the sense of optimization. I think we’re moving away from this kind of fatalistic view of dementia.

Professor David Llewellyn:

There will be challenges, of course. We’re going to see that the charities, Alzheimer’s Research UK, Alzheimer’s Society, who are two of the three funders of the DRI, they’re going to be struggling … Well, they are struggling at the moment, and that will be hitting dementia research funding worldwide.

Professor David Llewellyn:

So, we need to be realistic about that. I think we do need to just think about how we can do things cost effectively and collaboratively. And I think that that’s why there’s so much appetite to join the DEMON network and to do things in that kind of way.

Professor David Llewellyn:

And the fact that we’re over 600 members and we only launched in November last year, is just mind-blowing. I didn’t know if we’d end up being a little clique of 50, which was not what I wanted, of course, but I think it’s been wildly successful, given the scale of the investment we’ve put in.

Professor David Llewellyn:

So I would just say, if you’re a member of the DRI, or if you’re not, and you’re interested in these kind of issues, then go to demondementia.com and joining the DEMON network, and start to learn about the opportunities that there are. Start to meet with other people who have similar interests, or have completely different interests, but maybe there’s some overlap and maybe that’s where the innovation will come from.

Professor David Llewellyn:

So, it’s a great time to be doing this stuff. And again, I’m delighted that the DEMON network is officially partnered with the DRI. I think Bart’s vision about how to do things differently, it goes beyond just an ambition. What we’re doing is, we’re changing things in a practical way. We’re putting people together, and we’re going to get stuff done. So, what a time to be involved in dementia research.

Dr Megan O’Hare:

And you said you’re doing monthly seminars?

Professor David Llewellyn:

That’s right, yup. Every month, and every other month will be in collaboration, in partnership with the DRI, and will be relevant to all to the things that they’re doing.

Dr Megan O’Hare:

Okay. So maybe we can get a link for that or something and put it underneath the podcast.

Professor David Llewellyn:

That would be great.

Dr Megan O’Hare:

Yeah. Well, thank you both so much for today. We have profiles on both of our panellists today on the website. If you have anything that you’d like to add on this, please drop us a tweet using #UCLdementia. And, while I have your attention, I’d like to remind you that we have a great website, dementiaresearcher.nihr.ac.uk.

Dr Megan O’Hare:

Register today and you get our weekly updates, and we’ve got daily blogs, events and details of the latest funding cause, which is quite important at the moment, because it’s a bit changeable at the moment. So, thank you for listening.

Voice Over:

Brought to you by dementiaresearcher.nihr.ac.uk, in association with Alzheimer’s Research UK and Alzheimer’s Society, supporting early career dementia researchers across the world.

END