A Network of Networks
46:27 Share

You're listening to Data Skeptic, Graphs and Networks, the podcast exploring how the graph data structure has an impact in science, industry, and elsewhere. Welcome to another installment of Data Skeptic, Graphs and Networks. Today, I'm excited to get into a new area we haven't covered on the season, material science and the applications of network science there. Asaf, I believe you put me on the path of Benaiah to invite on the show. Tell me a little bit about your history knowing his work.

B'naiya was working with Professor Shlomo Havlin, who is a very famous scientist in the field of network science, got a lot published and many, many citations. I know Citation Network is very important to you. You have a...

Pet project on the subject, I think. Yep. We'll cover that probably more in a future episode, but very excited about how we can link people who've been on the show via their co-authors and things they publish with other people. Shlomo Hovland authored a lot of papers, and especially the work of physics and network science, and Benaiah was working with him.

They did an experiment, a very interesting experiment in their physics lab. The cool thing was it was the first time people tested not just networks, in this case network resilience,

but the connection between their networks in the real world. And it was, I think, the first time that someone tested the network of networks. What they did was looking at local failures in a network and see how it cascaded between the networks.

So from a network science point of view, it's unique not just because it's the first time a network of networks is tested, but also because they could focus on the microchanges in the network dynamics. Because it was in a lab, they could survey all the microchanges and they could see how the cascade happens in real time.

And they talk about it in the paper. I don't think you're mentioning it in the... I don't think he mentioned the cascading thing in the talk. We talk a little bit about the dynamics of failures. And what's kind of interesting in networks is how that can...

As you say, cascade or snowball. What seems like a small failure can really propagate the network and become disastrous. Yes, so that's what they studied. Now, Benaiah has left physics and went to Barabasi, the most famous network scientist, I believe, and is now studying network medicine.

The cool thing is that, you know, what's the connection between physics and health and so on? So the cool thing is that network science is very multidisciplinary. It's a multidisciplinary field because everything is a network. And we know from network science that all networks share the same laws. So we can apply the same logic. When Benaiah was talking about what he learned from physics and applied it into other networks, well, I

I personally am less interested in the physics application. I guess in this sense, I'm like most people. They either don't know anything about physics or say they know about physics and just plain lie about it. There's got to be a third category of those that actually know about it, though, right? I think it's so small that it doesn't count.

I'm more interested in the network perspective of their findings and the principle that you can model systems as a network and apply the same logic on them. As a network analyst, I use network laws and apply them on different networks all the time. So I'm kind of into it.

As I said, I won't lie, I'm not a physicist and I'm not so interested in superconductors, which Naya is probably much more interested than me. I'm more interested in the cascading in the network phenomena that was demonstrated by using superconductors. I think that's a very exciting area. I don't know so much about it, but it seems like material science could really be on the verge of blowing up quite a bit.

Network medicine is also an exciting area. I think there's huge opportunities in biotech in the coming century. So I was glad to see research like this taking place. Well, let's jump into this interview with Benaiah. My name is Benaiah Gross. I'm a Fulbright postdoctoral fellow at the Center of Complex Network Research.

at Northeastern University in the research group of Professor Lazlo Baralashi. And can you share a few details about what that research group focuses on? So what we do, we study many branches and aspects of network science. We study network medicine, how network affects diseases and drugs in your biological networks, in your body.

how networks affect physical networks, which is another section of the group, and many, many broad aspects of network science. Am I correct in saying your primary focus is around medicine now? Now I work on medicine. We'll probably talk about my previous work on physics application of networks, but both topics are very interesting.

For sure. Before we get into either, could you maybe talk a little bit about the nature of network science and how you're able to transfer those skills from semiconductors and material science into medicine, which to some people could seem unrelated? Yeah, so I think this is really the beauty of network science. So network science in its fundamental property

is that it's a multidisciplinary field. So network is basically a framework you use to study different problems and different systems. So you can use network science to study biology, like regulatory network or metabolic networks. You can use it to study the brain. You can use it to study social networks or infrastructures. So basically apply this framework of networks into different fields.

And the beauty of it, which we'll probably discuss today, is how you, that's how I see the research, is how you take ideas that develop in one field and you bring them to another field using the network framework. And I think this property is very, very strong. Could you talk about some of those insights from fields that transfer other places? The examples that we are going to speak about, I think, is the real, it's very, let's say, pure example of this transfer of ideas.

because we are going to talk about interdependent networks. Before 2010, every researcher was studying networks as isolated networks. In biology, people will start either metabolic network or regulatory network. Or when we talk about infrastructures, researchers will study transportation networks or communication networks and so on, but every researcher will focus on a specific network.

And in 2010, after a very big blackout that happened in Italy in 2003, there was a very big blackout in Italy. There was some local malfunction at the power grid, which eventually spread in almost the entire Italy, and the entirety was in blackout. And we talk about hospitals, everything.

And it was very puzzling for researchers how a local failure in a network can propagate data network. And it's not staying local. And in 2010, Professor Shlomo Havlin, which is one of the pioneers in network science in general, he said the following thing. When you look on a city, you have many networks. You have your water network, you have transportation, you have communication, many, many networks. They are not isolated. They depend on each other. They need resources for one another.

So the easy example is if you look on communication network and power grid. Communication networks need electricity to function. But also the power grid needs communication to make sure that there are not overloads in specific areas. So both networks depend on one another. And you can very easily conceptualize that if some part of the power grid malfunction, so there are some areas in the communication network that doesn't get the power it needs to work, so it shuts down.

So now when you look on the power grid back again, you start to have overload because there is no communication. So other part of the power grid shuts down. And so these failures, these dependencies between the networks can create failures to go back and forth between the network until the entire system collapses. This was the basic understanding that first, networks are not isolated. And these dependencies between them makes them vulnerable. So this was a very big breakthrough in 2010.

What we're going to talk about is how to apply this concept in physics. So it's basically ideas that was developed for infrastructures, for cities and the different networks in cities. And we take this concept and apply it in physical systems. And I think this was the main improvement and advances in the science of networks.

To wind up with that blackout, do you have a sense of the root cause of the problem? If I recall correctly, this was some tree that fell on some power station, and this was the source that initiated the entire failure. But it was not clear how a local failure can propagate. Why doesn't the failure stay local? To solve this, what Professor Havlin did in 2010, he basically introduced this new type of link.

Because when you think about a network, you have a network and the structure, the nodes and edges. The edges are what we call connectivity link. It's what structuralize the structure of the network. But when you have two networks that depend on each other, each network, the structure of each network is defined by the connectivity link between the nodes. And now you have dependencies between nodes in different networks. This new type of link introduce new mechanisms that allow failure to propagate

and basically propagate in the entire system. So as you'd mentioned, probably those networks were analyzed separately, like the communication network, maybe you've got a nice graph there and electric, a nice graph, but they're not connected in any way. Is the new idea that we should have some, I don't know, a hypergraph or just edges that connect them? What is the new formalism that better captures things?

You know that feeling when you discover just how much of your personal information is floating around online? I certainly do. When I got my first Delete Me privacy report, I was shocked to find over 700 listings containing my information across the internet. But here's the exciting part. Within just the first week, Delete Me had already removed almost 40 of those listings.

What really impressed me was their thorough communication. They keep you informed every step of the way. Sometimes data brokers like WhitePages need extra verification. DeleteMe lets you know to watch for those emails so you can complete the process. Plus, you get detailed privacy reports showing exactly which data brokers they've found your information on and what they're doing to remove it.

The best part? You're assigned a dedicated privacy advisor who's always ready to help with any questions or concerns. It's not just automated software. There's a real person watching out for your privacy. They continuously monitor for new data brokers and handle removals automatically as part of your membership. Take control of your data and keep your private life private.

by signing up for Delete.me. Now at a special discount for our listeners. Today get 20% off your Delete.me plan by texting DATA to 64000. That's DATA, D-A-T-A, to 64000. Message and data rates may apply. What is the new formalism that better captures things?

There's actually a lot that's been done in this direction, but the main idea is to basically, as I said, you introduce a new type of link. So you have your, if you do it mathematically, you have your agency metrics or some other way to formalize it, but you basically define the structure of the network and you define the interactions between the networks. You separate the two. And this was the main progress, as I said, was to understand that the interaction between the networks are very, very important.

Well, can you talk a little bit about how that insight can transfer into something like material science?

As I said, it was a breakthrough from the theoretical perspective, how to explain why failures propagate. But as a physicist or as a scientist in general, you want to be able to do experiments. It's very nice to have theory, it's very nice you can explain things, but as long as you cannot go to the lab and perform experiments and test it in different ways, it stays as a theory.

Unfortunately, or luckily, I will not destroy the entire power gate of a city every time I want to do experiments. So we needed to think about how we can apply this in closed lab experiments. And then it took almost over a decade, in 2023, it took over a decade to be able to actually do it in a physical lab. But we collaborated with Professor Aviat Friedman, which he is a solid state experimentalist.

And he worked on superconductivity.

So you can have physical networks, like superconducting networks. You can think about, again, a network, a node and edges. And this is specifically a very simple network. It's called the 2D square lattice. It's basically a square where each node is your node of the network and is connected to its four neighbors, the one above, below, and on the side. It's a two-dimensional network. And these networks, at very low temperature, it becomes superconducting. So each segment, each link becomes superconducting.

And if you are at a very high temperature, it becomes normal. It has some resistance. It sounds very simple, but it took time to really be able to do it. It took two networks like this and put them one on top of the other without them touching at all. When you study a single network, you start from very low temperature. The entire network is superconducting. So there is no resistance. The current flows without resistance. And then you start to heat the system slowly by slowly.

one by one, the segments start to change their state to have resistance until the entire network is resistant. In physics, we call it a second-order phase transition or a continuous phase transition.

But now what he did, he took two networks like this, put them one on top of the other, and he suddenly observed this jump. You start to hit the system, and suddenly by using very weak hitting, the entire system jumped immediately to the other state. When you think about it, it's exactly like the blackout. You do a small failure and the entire system breaks down, and here you do small hitting and the entire system moves from one state to the other.

And this was the first experimental evidence we could see that when you have two networks, you can observe these failures, these cascading failures. But what we were missing was what is the dependency? What makes the two networks depend on one another? And what we found was that the dependency here is realized by heat. So basically the state of each, as we said, the state of each edge or each segment depends on the heat of the system.

So when a single segment changes its state from superconductor to normal or resistor, it starts to dissipate heat by itself. It's called atomic dissipation. Now this heat goes to the other network. It heats another link, which also changes its state and starts to dissipate more heat, which goes back to the first network and so on and so forth. So its nodes start to change their state because of failure in the other network, and this keeps propagating until the entire system collapses or changes its state.

So one would hope, in theory, that two independent networks that just happen to be sandwiched on top of each other wouldn't interact. But I guess as you say, they're interacting via heat. Does that make the system in some way more complex to model? Right. So we obviously needed to adjust it to make more generalized theories. So when you think about

interdependent networks or network of networks, you need to take the old theory that looks on single networks, on the resilience and so on, and generalize it to many interacting networks.

to get more generalized theory and that's what we needed to do for superconductors. We took the regular or the standard theory for single network of superconductors and generalize it to many networks. In this case it was two because it's very hard to do experimentally but you can generalize it to many interactive networks and maybe in the future, this is one of the let's say future applications, is that you can start to think about what happens if I take many layers.

I take one, two, three, ten, a hundred layers and I can start to think about multi-layer materials. That if I can adjust how the heat is transferred between the two networks, I can basically create a material that responds to heat in the way that I want, according to whatever application I want to have.

So my naive thought here is perhaps you can link these two networks with a new type of edge that says, even though an electron cannot flow through this edge, you are my physical neighbor and maybe heat can flow. Is that an aspect of your modeling or you take a different approach? No, exactly. This is the exact point. You need to distinguish between the two types of link, between the connectivities that establish the structure of the network and the dependency between the networks.

So in superconducting networks, the connectivity within each network, they allow the current to flow inside the system. But as I said, the two networks, they do not touch each other. So current cannot flow between the network. The only thing that flows between the network is heat.

So this is very clear distinction between the two types of links. The main progress was to say different, you can establish new types of links that behave differently. I feel like that's a modeling choice. It's a more sophisticated way to describe the problem

Does that description allow the standard physics to come in and explain the heat transfer, or is there sort of new science to be done? Right. So now you can start to think about which materials, some substrate they put between the networks that can allow the heat to transfer, to have strong transfer or weak transfer. So there is a very big question about what is the distance, how far the dependency goes. Do the dependencies stay local, or does it go far?

One of the findings, which I think is very interesting, is that if you say that the dependency can go for long distances, you notice very interesting behavior. When the system breaks down, let's say if we talk about the blackout that happened, you have a single node that fails, which allows for a single node to fail in the other network, and so on and so forth. This is called a branching process.

which if you think about the COVID pandemic, everyone can really understand because during the pandemic, everyone was talking about the R parameter or R value, which basically tells you on average how many people with some infected person will infect. Critical value is one. In order for a pandemic to keep spreading and infected people

individual need to infect at least one another one if you have more than that it exponentially increase if you have less than one it goes down so as the interdependent network breaks down if the branching factor is exactly the same it's exactly one in order for the system to break down you need that a failure of a single node will on average make the failure of a single node as well and if

And if it's above than one, the system collapse very, very fast. If it's smaller than one, it will stay local. It will not spread. This understanding

allow us to develop a metric to characterize how close the system is to breakdown. So let's say you look on your city and you ask yourself, well, how do I know if the system is fragile, is close to the point where a local failure can propagate or not? How can I measure it? Every network has some failures. During the summer in my hometown, every summer there are overloads in the system and there are some local failures here and there.

So if you analyze these local failures and you can measure the branching factor, right, the branching factor has to be lower than one, right? Otherwise it will spread in the entire system. So, and you can measure how close it is to one.

is it 0.5? If it's 0.5, then you are so far from the critical point, you don't really need to worry. But if you measure it, you see that it's 0.8, 0.9, you need to start worrying. So this is a very important property to understand about how interdependent network works and how they can fail due to local failures.

If the Italian power system had a metric like that in 2003, do you think it would have highlighted the risk of a down tree taking down the whole network? In recent papers, this is what we're working on. You can identify how close you are to the critical point by measuring the local failure that happens in your system. But you can use them to say, look, we need to make the network more robust. You have to work with this or otherwise a catastrophic failure can happen.

Could you expand more on that metric? How do you calculate it and maybe what are its units and properties like that?

Right, so there are no really units. You basically need to... Sometimes you don't really have the data. Sometimes it's very hard to track. But let's assume you can track it. Usually you have some characteristic time or nodes are failing at a given rate. You can actually track this progress and make this analysis to, as I said, to measure the branching factor, which is very, let's say, simple measurement. It's simple parameters that can tell you on average if...

a local node that failed, how many other nodes did you fail? And theoretically, we started doing it theoretically first, but it's very straightforward to measure. And also in the superconducting system, we show that this is exactly what happened.

Exactly the critical points, the branching factor is one, a single segment, a single node make the change of a single node. And this is exactly the point you want to avoid. Or in other cases, if you want the system to collapse. So in superconductors, you might actually want the system to change the state. So you actually want to go to the state. But this is the point you need to look for and to adjust yourself according to it.

Well, in the superconductor case, it seems like, am I correct in saying you would want to control how the heat transfers or at least manage it more in the way you'd wish it would be working? This is very, let's say, application dependent. If you go to multilayer materials, as we said, it depends what you want the material to do, right? Do you want the material to be susceptible to local heat, for example, or not? It

It might be that you want your material to be able to propagate the heat between the layers. So you might, let's say, manufacture it in a specific way, or you might want it, don't want it. So if you can think about another application, which we often talk about, is about detectors. Detectors is a very big, a single photon detector is a very big thing that is very relevant, has many applications in many areas. The ability to detect a single photon,

But when you think about, again, the interdependent networks, it's exactly, let's say, a straightforward application. You want a single photon to hit your network and do this local heating that will spread in the entire network, and this will basically detect the photon. So that is what's very nice for having a laboratory setup that you can tune and adjust and perform experiments to really validate the theories and build some applications, depending on whatever you want to do.

Well, I know when dealing with heat, we're sort of always headed at the increase in entropy. I'm also reminded of Maxwell's demon that, you know, we can't open this door and contain all the heat in one side of the room or anything. Do you face the challenges of, you know, the physical nature of the universe in this work?

we didn't tackle it enough in this direction. So this is a very good question and this is where we will go to ask these questions, which so far, as I said, it's a very new result that we have. And so far we just established that you can see the phenomena, you can understand it. But now if we want to go deeper, this is exactly the future work that we are doing. But I would like to point out two points.

The first one is that the dependency between the layers is by heat. But you can start to think about, if you think about the future of this research, is can you create dependency between networks by other physical properties? Can you use magnetic field? Can you use pressure? You can think about many, many other topics. You can also think about, can I use other materials? Can I use magnetic materials? Can I use elastic materials?

The main achievement of this research, in my opinion, as we go back to the beginning of our talk, is the ability to transfer ideas between different fields of network science. Because this is what we did, right? We took an idea from infrastructures and applied it to physics.

What will be the next idea that you can transfer between networks? This is the real thing that we need to think about. We still work on, as I said, on superconducts or on materials and all of this, but where will be the next breakthrough? The next breakthrough will be when you can think about another idea from one of the fields that network science is applicable and bring it to physics and bring it to another field. This will be the next breakthrough.

Are you able to take the known laws of physics and just apply them, or do you have to come up with new models and new science in that style of work? When you think about the concept of interdependent networks or networks of networks, the beauty of it was that you take the existing theory of a single network or a single system and you generalize it.

It means you don't invent something new by itself. You basically take the existing topic and say, look, you just looked on a specific case. A single network that does not interact with any other network is just a specific case. The most generalized case is that many networks interact with one another. It's not that you invent a new law of physics. You take the existing law of physics and you say, look, you just looked on a specific case. It's not the most generalized one. You generalize the existing theory.

And by generalizing it, you discover a new phenomenon. In this superconducting case, do you have a sense of where it goes? It's good enough to do research for the purpose of just science and basic science, but do you have any idea about the industrial applications? What do we get if we can effectively layer superconductors and control the heat transfer?

There are two main applications that we talk about. The first one is single photon detectors, right? The ability to detect a single photon. Well, let's say, maybe I didn't emphasize why exactly it's important for detectors. So let's assume you take a single network, not interdependent, a single network, and a single photon is hitting the system. So it induces some heat, but because the transition is continuous,

you will not be able to distinguish between this heat and some noise you have in the system. But if you build your interdependent network of superconductors and adjust the parameter in the lab as you want, you can make sure you can at least try, this is one of the applications we're working on, be able to distinguish between the impact of noise and the impact of a photon that hits your system. And this will be

you will be able to use as detector. So this is one application. The other application, as I said, is the multi-layer material. You want to basically start to think about stacking many, many layers on top of the other and adjust the parameters between the layers in a way that you can construct the material with the properties and the behavior you want. One of the things we sometimes talk about is anisotropic material.

You can think about materials that if you hit it in one direction, it will not respond. And if you hit from the other direction, it will respond. And the way you build this, the way you construct the interaction between your layer, the interaction between the network can adjust the properties of the material towards this direction.

So in the case of your closed lab experiment, you have an idealized case. The rest of the world is isolated, or at least within reason. For something like power or supply chain or any of these real-world networks, you don't have the luxury. There's no closed experiment. Can the modeling still apply there?

Yes, yes, at least in theory, right? I don't think that researchers got really deep down into the really deep analyzing of each of the networks because there is a very nice, let's say, visualization of it when you can actually show in a city how the different networks, how their topology is different, how exactly the interaction between them, you have a very big picture of network of interacting networks that

interact together and then if you really can classify the exact properties of each one and the exact structure and dependence between them and you can also start to take into account noise and local effects then you can really have good predictions but i think having this let's say deep down understanding and of the data and the exact structure of each network is not yet available

And you talked a little bit about where places you want to take that insight. Given what you've learned in the lab, what are some of the domains you hope to apply these sorts of approaches to?

First of all, the material world is a world that we are not there yet, right? Because what we did was only two layers. Two layers are not a material. It's just two layers. But this is the future of this project. Eventually, we'll go to material science. This, I think, can really revolutionize some aspects in material science that will basically allow you to create multilayer interdependent materials with the properties of your desire. I think the more bigger picture

is how to get new ideas to transfer between networks, right? So we didn't talk about this, but after this paradigm was established, it was also found to be applicable in biological systems. So you have your metabolic network and regulatory networks, and what you find is that there are some dependencies between the networks.

So it means that so far, researchers that look only on metabolic network or only on regulatory network had basically a partial picture. Also, when you go to ecological systems, right, so you have species, you have planets, plants, and you have some interaction between them, what you really have is a network of ecological networks, because a plant is not interacting only with a specific species, interacting with others. I mean, you build the entire

interaction of your ecosystem, what you notice is it's an interdependent ecological network. So what you start to find, I think this is taking an idea and applying it in different fields of network science. But the future will be, what will be the new idea? What will be the next idea that can be applicable in all the other fields of network science?

What about specifically within network medicine research? Are there areas you can apply some of the learnings from the lab experiment there? Right. So this is a very good question. So maybe we should speak about, slightly about network medicine. Until 2007, it was a network biology. That's how researchers call it. So you analyze network in your body. It can be gene network, it can be metabolic network and so on.

And in 2007, Professor Laszlo Barabaschi, which is also one of the leaders of network science, he said, let's look on how diseases are established in these networks and medicine in general, some application in medicine. And what he found was a very important breakthrough. He found that diseases are not random. Okay, let's go back a little bit. When you have a disease,

you have genes that are associated with it. Sometimes when you have a specific mutation, it creates a disease. But in other cases, let's say Alzheimer's disease or other diseases, you have genes that participate in the mechanism of the disease or genes that are highly expressed for individuals that have this disease. When you look on the human interactor, which is the network that represents all the biological interaction in the human cell,

You look on this network and you ask yourself, well, if I look on a specific disease, where are the genes that are associated with this disease? Where are these genes in the network? And what he found was that the disease genes are not sprained randomly. They agglomerate in a specific network neighborhood. This is what, and basically they form what we call a disease model. It's basically a subcluster that represents the disease.

And what was very interesting about this is that afterwards you can actually map the diseases, where each disease is placed in the human interactome. And then you can start to see the relation between diseases. You can see that diseases that behave similarly are closer in the interactome than diseases that are separated.

You can see where the cancer diseases are, you see when you have hypertension. So many, many different diseases, you can actually have a map of diseases. And then you can ask yourself, now that I want to look about drugs, I want to see which drug affects which disease, I can use the network of diseases to take

that was built for one disease, was created for one disease and use it to affect another. But this was also done during COVID, right? When we had COVID, we didn't have a cure for it.

We didn't have a medicine. But so at the beginning, they used to take a medicine that was developed for another disease and use it for COVID. And it was actually, it worked many times. But when you look from network perspective, you can look on how the drugs that were developed for one disease, this disease is very close to another one. So there is a very high chance that this drug will also be useful for this disease. If you think about interdependent networks,

And you can ask yourself, well, how this concept of network medicine will work if I look on interdependent biological networks, right? So you can build your interdependent metabolic and regulatory networks. And again, ask yourself from the beginning, all the fundamental concepts of network medicine, how they are established or how do we see them now in this more general structure of interdependent networks?

When I think of the genome, and perhaps this is the naivete of my understanding, it's the four base pairs and they're in some sequence. So it's sort of like a linked list data structure to me, not so much a graph, even though obviously a graph is a general thing. A special case could just be a sequence. Is there more to it that is graph and network like than just a sequence?

Yes. So the human interactome is a network that represents all biological interactions in the human cell. So you take all the proteins in your cell, right? And you see which two proteins interact. And they can interact in different ways. But if they interact, you build this edge between them.

And it's a very big work. So the current human interactome, if I recall correctly, it has over 18,000 genes and over 1 million interactions between them. And you have this very dense, complex network that represents all interaction. If you include also RNAs, so it's even more complicated, but you can include all the human interaction in the cell.

And once you have, this is the basic network in network medicine. And you look on this network and then you ask yourself all sorts of questions, right? What are the diseases in this network? What are the drugs? And so on. How they interact with one another. There are also new developments now about the food space, right? You eat some food, right? Which is built from food compound. And these compounds go and bind to specific proteins.

in the human interactome and you can start to analyze how your diet is affecting different diseases and so on. So there are lots that can be done about health in general. I specifically now applying this to aging. This is the network we are looking at and we start out to see how the different nodes and different properties of the network can actually be applicable or can transfer to our health.

When it comes to aging specifically, I guess the first step is understanding. Do we know its mechanisms and that sort of the science? And then perhaps later, can we affect it or change it or stop it? Maybe that's a bit science fiction. I don't know. Where do you feel we are at, you know, sort of the royal we, where are things at on the journey to understanding aging and maybe controlling it to some degree?

When you talk about aging, first, aging is a very big space. There is a lot of work that's been done about aging. But when you talk about aging, you first ask yourself, well, what is aging? At first, it looks like a weird question. I mean, what is aging? We all know what aging is, right? But in this space, you need to distinguish between why do we age and how do we age? When you ask yourself, why do we age? This is what we call the theories of aging.

And currently in the space of aging, in the world of aging, there are three main theories of aging. The first one is about accumulation of mutations, right? So as you age, you accumulate mutations, you accumulate damage, right? Your networks, your bio-aging networks and biological networks, they start to get damaged. And this damage basically is the reason you age. Your cells start to function, let's say, worse and worse. And this is aging. This is the first one.

The second theory talks about epigenetics. So it's not damage, but rather genes cannot express themselves in the right way because of mutilation, because of stomatification. There are some different reasons, but eventually it's not a damage itself, but rather some, let's say, overlay, over layers that don't allow the gene to express itself, which if this is the correct theory, it's very good news, right? Because you can then

let's say, reprogram the cell back, because it's not damaged, you just need to take it back to the younger state. And this is actually, there is very good evidence for it, which is called the Yamanaka factors, which is basically a few transcription factors that can rejuvenate the epigenetic state of the cell, and you get a young cell back again. So this is very breakthrough in recent years.

The third theory talks about, let's say, accumulation of waste. So over time, when you have metabolic processes, right, you eat your food, you have metabolic processes that eventually gives the energy to the cell and what the cell needs.

you have some leftovers, right? Because when you eat something, it's not exactly what the cell needs. So there is a mechanism in the cells that take the waste and disperse it. But over time, it's not working as good as it should have. And you start to accumulate this waste inside the cells that stop the function. So this is usually the main three theories of why do we age. But when you think about how we age, this is slightly different topics. This is what we call the hallmarks of aging.

So there are, at the beginning it was nine hallmarks or 11 hallmarks, but it's nine or 11 hallmarks, biological processes that happens as you age. You probably heard about your telomeres are getting shorter, right? So the epigenetic changes, you start to accumulate cell senescence. These are the processes that happens as you age. And once we looked on the aging in the perspective of network medicine, what we found is that each of the hallmarks of aging

has a hallmark model. Aging genes are not random. They are located in specific areas in your network, in your human interactome network. And not only that each of the hallmarks of aging has a specific cluster, all of them are in the same network neighborhood. So they are one next to the other. And all of them together is what we call the longevity model. This is the model. This is the area that represents aging in the

human cell. And once you characterize it, you can start to think about, can I find drugs that affect this module?

So you can look on the target or you can what you usually do, you go to the drug bank, which is a very big repository of all the existing drugs. It has over 6000 drugs. And you start to look on drugs, which their targets is very close to the longevity model or hitting the longevity model.

And once you do this analysis, you can have a short list of drugs that you can recommend for experimental validation that might have the chance to affect longevity.

Do you deal with any engineering challenges in your work? I imagine some of these networks get large and difficult to calculate with. When you try to do drug repurposing, you want to use your network to take drugs that were developed for a specific reason and to see if it can be useful for another disease. So when you do that, you need to scan the entire drug bank, which, as I said, has over 6,000 drugs.

and to scan all the diseases that you have, which is also a big repository, right? Each disease will have the genes that are associated with it and start to find connection between them. It does take some time. So when you write your code for this analysis, it can take two weeks or three weeks. But I didn't really go to a position where I needed to have like a very, you know, unique solution of using different machines and different

very unique infrastructure or using GPU, for example, I didn't really reach this limit. But we might, as the field gets more complex and the analysis become more complex, we might need it. So for example, something I didn't mention earlier,

Even if you find that the drug will hit, you do your analysis and then you find that the drug will hit the disease that you want. How do you know it will be useful and may not harm it? So if you talk about aging, how do you know it will delay aging and not accelerate aging? So this is something we work on because it's very important to distinguish between the two. So this is another level of analysis.

that now join the pipeline of network medicine development. So over time, I believe, other parts will join. So you need to start, one of the things that you will think about is side effects, toxicity, and many, many other, let's say, some aspects of drug development that need to be addressed. So over time, this will join the pipeline, and eventually you might get to a position where the pipeline is so complex and computationally hard to do that you will need to use some complex engineering way

to do it right. But currently we are not there yet. Well, we've touched on this a bit, but I always like to ask guests, what's next for you? I've been working on the aging space for a year and a half now. And I mentioned it slightly a little bit, but we're going out to the food space.

So we are talking mainly about drugs, right? You want to take a drug that was developed for one aim, right? For one disease and to use it for another, but you can start to think in the same way about food, right? Again, you eat food. Food is, say, you eat apple, right? Or you eat the garlic, whatever. You have many, many compounds inside it that are not

Let's say you don't read about it when you look about the nutritions in your food, right? When you talk about nutrition, usually it's about vitamins and very simple things that we can understand. But there are many, many other compounds or other molecules inside your food that are much less studied. This is what we call the dark matter, nutrient dark matter. It's not the dark matter of physics. It's another one. But when you look on this compound... But we don't know, right? Yeah.

But when you look at these compounds, what you notice, some of them behave like drugs. Actually, many drugs that you take for diseases are actually food compounds that were synthesized.

So these compounds that are inside the food you eat, they also bind to proteins. They also affect your network. And then you ask yourself, well, can I characterize the compounds that affect the disease model or affect the aging model that I study? And then I can actually recommend a diet. I can tell you which food you need to eat to contain the food molecules that bind to the disease model to affect it in the right way.

This is kind of a relatively new space in network medicine, but let's say not in the near future, maybe more in the far future, you will have diets that will tell you which food in your diet has the specific compound to affect specific disease or phenotype in your network. And then you will have a network-based diet.

So this is kind of, let's say, a future case that we'll have. Dai, what is the best place for listeners to go online to follow you and the lab and anything else you have going on?

Okay, well, I'm actually not so, let's say, I want to go to conferences. I'm going to the APS conference soon in March in California to talk about interdependent networks. So if some of the listeners are going there, I'll be happy to speak with them. I would recommend, well, go to Laszlo Barabasi's website. I work in his lab.

And obviously he's very big in network science and he has many lectures and many topics and he does very interesting work in all aspects of network science in general. So, and in general, if you are in Boston and want to come and chat about different topics of networks, you are welcome. Very cool. We'll have links to the show notes for all of the above. Thank you so much for taking the time to come on and chat about your work. Sure. It was my pleasure. Thank you for having me.