248 – PBDM and Public Health: Bridging Modelling Across Ecosystems and Epidemiology

[00:00:06] George: Hello and welcome to the IDEMS Podcast. I’m George Simmons and I’m joined today by David Stern.

Hi David. How are you doing?

[00:00:13] David: I’m good. Looking forward to discussing more modelling.

[00:00:18] George: Yeah, in a previous conversation we talked about our immediate plans with this PBDM approach, and the modelling work that we’re doing around that. In this conversation, we’re gonna, I guess, discuss something adjacent, but from a slightly different angle, where we want to think about these ideas in the context of public health. And there’s been lots of little developments that I’ve had and you’ve had, so part of this is just to update and discuss everything that’s been going on.

So I’m happy to start on my side. So there’s been, I guess, two different main things that have happened recently. The first is actually a workshop I went to in the US hosted at the American Institute of Mathematics. And this was about thinking about formal modelling, which is what some people call the category theory kind of modelling frameworks that we’re doing in the context of epidemiology.

So the idea there was to gather experts in categorical modelling and experts in epidemiology, throw them all together and generate ideas on how compositional modelling approaches could serve epidemiology, and also how epidemiology and the current research there could inform what needs to be developed on the mathematical side.

This is very interesting ’cause this is kind of the first time for me that I’ve been exposed to the kind of modelling and the style of modelling, this formal framework, in a context outside of the traditional agriculture and agroecology context that we’re doing. 

[00:01:53] David: When you say traditional, it is only us really who are doing it.

[00:01:57] George: Traditional for me, exactly. And yeah, one of the big realisations is that, in some ways, the issues are very similar and in some ways the issues are very different. The ways in which they’re similar is that people are in all cases trying to think about how to build bigger, more holistic, more representative models out of bits and pieces like smaller models that represent certain things.

So, in that context, it was community disease transmission, traditional epidemiology models, a recovery model and a public health intervention model. How do you combine those things together in a compatible way to create something that is holistic, is useful, and can actually be interrogated for good answers.

The ways in which they’re different was very interesting. One of the ways which I found really interesting is that the actual way that the modelling dynamics function in what we do and in what epidemiology does is quite different, and part of the reason why we’ve spent quite a lot of time actually developing our modelling system is to be able to generate much more complex transition rates.

Often in agricultural modelling, the rate at which things happen are very much not constant. And they can be functions of the climate, they can be functions of internal states, they can be quite complex kinds of allocation processes within a plant. And what we’ve spent so much time trying to get hold of and make accessible is how you can construct those things in a nice, easy way. So there’s this really interesting balance that I came across in that context.

The second thing that I’ve been up to is, you put me in touch with a professor at the University of Johannesburg, he is also looking at epidemiological modelling, but in the context of wanting to construct something more holistic, something more system wide. Some of his current work, for example, looks at how you combine two scales of modelling. One is the model of how a disease progresses within the kind of transmission vector or the disease host, and the other is how the transmission occurs between those two.

So there’s two very different timescales going on in the context of that modelling. And part of his vision is to actually understand how to keep going, keep adding realism, how to make that disease vector, which often is an insect, for example, be a more representative model. And I believe part of why we had that conversation was to try and connect our work in ecosystem modelling with that kind of epidemiological modelling where you are considering a lot of things across different scales, spatial resolutions, timescales, and so on.

The other interesting thing about that conversation obviously came from realising the similarities in again that way that people are trying to build bigger models out of smaller models. Even the kind of composing of the within host or within vector disease models with the transition models is still a type of composition.

So there are definitely two levels I see that we share similarities and could pursue work. One is this quite deep structural thing where we could understand what that kind of cross scale composition is. And the other is building up a useful or more system wide representation by trying to link epidemiology and public health angles into our PBDM approach.

[00:06:04] David: This is great because I set you off on that meeting knowing that it would be useful, but I was interested to see what would happen. And it’s brilliant because I’m really conscious that the heart of what we’re doing and the approach we’re taking with PBDM is not unique to that context. And to hear you making those links of seeing how, ah, okay, this is why we’ve got this level of generality, because broadly the approach we are talking about is really the same but different.

It’s a different context, it’s a different set of parameters and models which are getting built. But it is the same basic principle of how we’re wanting to break down into components and look at the composition of these components, looking at different scales and how they can then be composed together and how we can do so in this way, which enables us to work across specialisms and across experts in different contexts, to build that collaboration of people working in these different areas.

And that’s exactly the same. And it’s really nice to be discussing with experts in a domain where we are not experts yet, and to see that actually, oh, that’s where you are at, but I already know how to do some of the bits you don’t know how to do, and you might not even have seen yet that this is what you need to do.

A lot of the big advances in science come from crossing disciplines, crossing these boundaries and seeing the similarities. And so it’s really great to see that understanding that what we’ve started to build has a particular application that it is built around, but it applies so much more generally. And I’m really excited about the entry into this in a public health context, be this epidemiology or other elements.

And the within host disease modelling is actually very well established and very interesting. There’s a lot of work which has happened there, and so to start to see how these different things can be brought together and can relate to one another is very exciting.

[00:08:45] George: Yeah, I agree. And we can get into a bit more about what we see specifically with PBDM and public health. But, I suppose, it’s actually a third level that you brought up, which is this enablement of collaboration between modelling communities. That piece is really linking to the work we want to do with people like the Topos Institute, where they are thinking about how to better enable those cross-disciplinary interactions in a way that help so many more people get started in them, who very often might be shy or think it’s too resource intensive to start having those kind of conversations. Things like having different translations between modelling paradigms, unified frameworks, all of those kinds of things are a third element to all of this.

And this came up at the AIM workshop. Ultimately, people modelling very different things at the very basic level use the same toolkit. This is kind of the core thesis of why you can enable these kinds of collaborations, because when you undress all of that technical language and these simulation programmes that are wrapped up in solvers and very complex lines of code and whatever, when you actually unwrap everything, it is the same toolkit underneath. What differs is how people communicate, how people think about it. And yeah, I think that that third layer, embarking on these public health ideas is really interesting.

[00:10:22] David: Absolutely. It’s the great privilege that I feel that we have. And this is something where not enough mathematicians, I feel, appreciate this and value this, but this privilege of being able to cut across disciplines and find the commonalities is in some sense what I feel as a mathematician I was trained for.

[00:10:45] George: Yeah. 

[00:10:46] David: To get to that core, what it really is about, was a big part of my training as a pure mathematician, I’m sure you feel similarly. And I’m amazed at how so many of these other areas are just not exposed to people with this sort of skillset because mathematicians don’t tend to engage in a way which is of service or in service of other disciplines. And it’s so much fun to be able to have these interactions, to actually understand what people are trying to solve and why, and how, actually, yes, some of the things which people are struggling with, well, there are real solutions to this, which are possible.

I come back to the PBDM approach and Andrew’s amazing work on this because he is so mathematical in what he has done in actually recognising the essence of what matters and this very overused phrase:“all models are wrong, but some models are useful”. Actually, the heart of that is, if you’re going to get into modelling things, what is actually useful? And as mathematicians, this is the essence of the attitude that I love to push for, that we just need to be useful.

[00:12:12] George: Yeah. And these are ideas that we’ve tried to communicate before, there is that innate training, whether you are trying to find links between combinatorics and topology or people modelling disease transmission and ecosystem scientists. There’s the same kind of skillset of understanding what’s actually going on underneath.

I’d be interested to go into a bit more of those other levels. This idea that we can actually start to think about now, given our progress on the PBDM modelling work, of actually trying to attach those models to disease transmission models. I think this is a really exciting avenue. It’s not something we’ve attempted to do, but we know that there is appetite for this.

[00:13:05] David: Absolutely. I want to actually just take a step back to the agricultural context for a second, to explain why Andrew’s work is so foundational on this and so powerful. If you take the agricultural context where we have pests and diseases, but we also have crops, then the fact that you could have a more complex, detailed crop model, which is part of, and fits into, the PBDM approach – where essentially what you are exposing is just the interactions of how that model will then interface with the insects, let’s say – is really easy for us to imagine.

This isn’t what they have done in the past with these implementations, but it is something which we already put together, and unfortunately it wasn’t successful, in an EU proposal, which would’ve done exactly that. And it was really easy for us to imagine and see how that would happen.

But that’s the same basic principle as for these epidemiological models or the disease models within the host. The power of Andrew’s PBDM approach is that you can actually, in the category theory approach, black box and just worry about the ports, the interaction points. And what you are actually interacting with can take many different shapes. So this idea of integrating very different modelling types into a coherent system is so natural, it’s really easy to see. This is not conceptually hard at all from the basis of what we’ve already built.

[00:14:57] George: Yeah, absolutely, it’s that kind of core conceptual level that we keep coming back to, that these basic tools, the frameworks that different groups use, they are the same. And we are in a place where we have those kinds of systems where we can actually sit down and try and communicate with some of those other different pieces of modelling. And I think that’s a really exciting avenue that I think we’re going to look into building up.

[00:15:28] David: There is actually a group that I’m really looking forward to engaging with, and this will be another episode we’ll have in a few weeks. There are people who have been looking at the same idea from a very different angle. We’ll leave that for another episode, but it’s really exciting that there do seem to be a number of pieces of the puzzle which are converging, where different pieces of work that have happened have reached a level of maturity. 

Andrew’s been working on this for 50 years and it’s incredible to me that the ideas that he’s been publishing on so widely, bringing up consistently for 50 years haven’t been picked up more widely. But I understand why, that until a few years ago, the technology to actually implement this always felt out of reach. He was able to do this on a case by case basis really efficiently himself, but even just building teams of people to do this without him was challenging because it required his intuition, his skills. Building those tools was not easily achievable.

Whereas there are some recent advances, both mathematical and in terms of how models are implemented, which I believe now mean that the solving piece of this is trivial. You are not building any solvers, you are just using existing solvers in everything you are building. So that piece of the puzzle is no longer something which is a barrier. Now there are issues about optimization of how you use solvers in ways which actually make them efficient enough to compete. That’s gonna be hard. But the point is that that’s an optimization problem, not an implementation problem, and so it’s separate.

The same, I believe, is going to be true in this sort of public health sphere. The discussions in that area are such that a lot of the mathematics that’s needed to implement these is actually known now. But it’s not the case that that’s always easily accessible to the public health, the medical, epidemiological researchers, who are faced with problems at a given point in time.

So bringing these things together requires a different mindset, which again, this is where the Topos work is starting in that direction, but they’re coming at it by saying “what’s the foundational problem?” Whereas I think what we can do in a way which is different, is actually say “well, let’s not worry about the foundations right now, what is your problem?” And actually take it problem by problem for so many individual problems where we then build the links into the foundational problems. So actually start by looking at this with some very concrete cases.

I was discussing malaria recently, and this is a beautiful case where actually there’s been a lot of very good work on modelling malaria, but at the same time, there’s a lot that could still be done and there’s a lot that needs to be done because it’s such an important public health concern.

[00:18:46] George: Yeah, I think one of the points you made there is really important, which is, you know, going back to Andrew’s context and this PBDM, part of the reason it doesn’t catch on is because people who are decision makers around agriculture do not believe it is possible to construct such models that he does. And it’s the same here, it is that mindset issue of how do you shift mentality to say it is possible to construct bigger things, it is possible to model this in a way which is more representative, in which we can combine this model of malaria with this model of mosquito and we can actually start to answer those questions. 

[00:19:30] David: Let’s be clear here that there’s plenty of people who have used modelling, and in some sense oversold modelling, in terms of what it can do and what it can be used for, for decision making. But when you dig into what those models are, they’re not representing the underlying phenomena. And at the heart of Andrew’s approach is that you model the individual phenomena for what they are, and then you put them together.

And so you are not trying to do things which are extrapolating, you are trying to understand and to model the underlying phenomena which are actually happening. And this has been going on in physics for years in these big physics climate models. But to do this for biology, that’s been something which, as you say, Andrew’s been doing in specific context with some of his collaborators, and in most other contexts it’s been deemed too hard.

[00:20:31] George: Exactly.

[00:20:32] David: And I think this is not a computational issue, it is a conceptual issue, although there are, of course, computational limitations.

[00:20:42] George: Yeah. But the work Andrew has done, he’s managed to run models of this scale for 30, 40 years, it is not a computational issue to be able to understand an ecosystem as a whole. It is, as you say, very conceptual. And a lot of the work we need to do, and a lot of the challenges we have also had in writing the grants that we mentioned and trying to talk to people, is that trying to break down that conceptual barrier of what is or what is not possible, and, for us, more often what we believe is possible. 

[00:21:21] David: And I think we’ve got to be careful not to try and oversell as well. But the thing which I think we can do, and we need to do, is really to consistently communicate that getting models to be able to be built from understanding the underlying phenomena, be it biological phenomena or chemical or physics based phenomena, actually starting with such building blocks to be able to build and understand systems, is possible. And that that is an approach which is likely to draw out insights that you will not be able to get unless you can model those phenomena. This is what it comes back down to.

I think it’s worth finishing with an analogy maybe, where in Andrew’s early work, he demonstrated that there was an insect that people were using a lot of pesticides, were using planes to pour pesticides all over the place to get rid of an insect.

[00:22:41] George: On cotton this was.

[00:22:42] David: On cotton, yes, and which was not actually affecting the production of cotton. 

[00:22:49] George: No, and it was actually probably positively affecting.

[00:22:52] David: Yes, it was. Probably, leaving it would have positively affected the production of cotton. And so this saved huge amounts of money and it reduced the negative impacts of the pesticide, environmentally as well, and so on.

And think of the parallel in public health. In COVID there were real questions about the lockdowns, and how important these were, which I don’t disagree with, but they were a very blunt tool, a blunt instrument because we couldn’t produce anything that was better. And part of that was because there wasn’t really the possibility to understand what a better tool could have been. Imagine in the future, instead of having such blunt tools, such as lockdown, we could actually understand, well, for a given disease, what would be a more effective way to limit spread?

[00:23:59] George: Yes. Is it closing the schools? Is it stopping people using trains?

[00:24:02] David: Exactly.

[00:24:03] George: Yeah.

[00:24:04] David: What is it that’s really needed? What will actually happen? Could we actually compare such things because we have an understanding, and how this changes from one disease to another?

This is not a criticism at all of public health policy. It’s just a statement about where we are as a society in terms of being able to understand. Can we have a better understanding of what is needed and what is actually effective and efficient by understanding the systems better, and actually being able to understand how things work within systems?

This is something where, now, I am not trying to promise, you know, if somebody were to ask us to build the models that would help protect against the next pandemic, we’re not there yet, but can we actually get so that scientists could work together to build the underlying models that could enable as a society to work towards that? That’s what I think there could be a large collaborative effort, and it’s how we can actually get the efforts that are happening to be more coordinated and collaborative, to be able to contribute towards systems that would be able to be used in such ways.

That’s what I think we can start to do. It’s exciting, these are important things.

[00:25:30] George: Yeah. That’s a great note to leave it on.

Thanks David for your time today. 

[00:25:33] David: Thank you.