 this miniature landscape encompasses a lot of what I've focused on during my PhD. So obviously it does this in a literal way, since my work has centered on microbes. But as we'll see, and I'm sure many of you know, the microscopic world is teaming with incredibly complex relationships. And that's why I also like to think of this picture in sort of a figurative sense, since what my work aims to do is find a way to navigate these complex landscapes, if you will. So what I'll try to do in this talk is sort of go through some work I've done over the past few years, which I hope will also paint a picture into my own journey into complexity research. So to start with the basics, why I guess do we even care about microbes in the first place, right? Or could possibly be so complex about them, you might find yourself asking. And for me, it's a little easier to get a grasp on their impact just by thinking about how long they've been around. So this right here is a geological summary of life on earth, which I think is a really simple but kind of great tool for putting this into perspective. So microbes have been around for around 80% of Earth's history, whereas we have only been around for 0.004% of that time. So they've been here for really long time and had a lot of impact in shaping what the planet looks like currently. And they continue to do a lot of stuff. So they're integral to the cycling of nutrients in the soil and in the oceans. And they're really important to our own health, where we have learned more and more how the trillions of microbial cells associated with the human body are crucial to things ranging from digestive processes to immune regulation. So I guess all of this is to say that we very much live in a microbial world. And it's therefore really important to understand the different types of things that these organisms are doing. So different types of microbes, I should say don't just live on their own. As many of you probably know, if we zoom in on any one of these ecosystems, we would obviously see that they live together with a bunch of different species making up microbial communities. And as such, the types of microbes that make up a community and what they're doing is really important. So a classic example of this is in the intestinal microbiota, where a certain composition of microbes can be associated with a healthy gut. Whereas a different composition of microbes can be associated with like an inflamed or disease state. So this question of who's there in terms of the microbes that are present is really important for the health of an ecosystem. And in recent years, we have been able to address this question of who's there in a variety of biomes like in oceans around the world, in soil ecosystems, and on a bunch of different sites of the human body. But knowing the composition of a community is only part of the story. So knowing what the microbes are doing and how they're working together or interacting in these ecosystems is also really crucial. So then how do we describe these microbial interactions? Well, for example, if we have two microbes, and one is making a particular molecule for another, we can call this commensalism. If the other is reciprocating with a different molecule, we can call it mutualism. And both of these would be examples of cooperation. But organisms can obviously also compete for particular nutrients. And these outcomes can let us classify interactions in different ways. But this framing is still fairly simplistic and definitely has its limitations. So for one, looking at interactions only in terms of their cooperative or competitive outcome starts to become very difficult when you have a large number of species. So as an example, this right here is a correlation network that we can use to infer interactions in the human oral microbiome, where every node here is a particular microbial genus. So for complex communities, like the ones we encounter in nature, understanding how all of these interactions take shape can be a really challenging task. And I think this is encompassed really nicely by this, which is one of my favorite tweets, which says, let's not think about more than two microbes interacting at a time and everything will be okay. And even to add more complexity to this, it's not just the communities can contain hundreds of interacting species, but also that these interactions themselves depend on a bunch of different factors. So how the organisms are arranged in space impacts how they exchange nutrients. There are, for example, bioluminescent bacteria and squid that glow based on the day-night cycle, so temporally dependent interactions. It can be metabolically costly for the organisms to produce exchange molecules. And then other factors such as pH, nutrient availability, and the temperature of the environment can all influence how these microbes interact, and therefore what the community will look and behave like. So I guess all of this is to say that interactions are these complex and multi-dimensional phenomena. And this complexity makes them hard to study, but the ability to know what they're doing will let us harness them to do a bunch of really helpful things. So this is what's called, I guess, the new field of synthetic ecology where people try to engineer new microbial communities for applications in health and the environment, among other things. So just to throw off a few examples, there's a number of companies right now making microbiome therapeutics. We're starting to understand how the microbiome modulates our immune response. And some applications have even reached clinical trials such as this development of treatments for Crohn's disease. And outside the area of human health, these microbial communities are relevant to the production of new biofuels, can be engineered to clean up oil spills, and finally there's a lot of work that has been going on to make microbiome treatments to increase crop yield and make them more resilient to disease and environmental pressures. But I guess in order to reach this promise, a lot more work definitely has to be done to make sense of the complexity of microbial interactions, as well as all of the different factors that I mentioned that can influence communities. So this will bring us to the first topic that I'll discuss today, which deals with a framework that we designed to kind of help us make sense of the different factors that influence these microbial interactions. So, as we now know, high throughput technologies, like for example 16S sequencing, metagenomics, metabolomics have allowed us to learn a lot of what's going on in microbial communities. And of course, the vast majority of these studies focus just on a single biological system or experimental setup, which lets us look really deeply into those specific conditions, but limits our ability to compare interactions across different studies and infer larger scale patterns. And having this ability would let us start tackling really big ecological questions. For example, are certain kinds of interactions more prevalent in certain taxa, or even how common is mutualism across all biomes? So what we need, I guess, is a way to somehow integrate all this data on interactions that we've been collecting, so we can systematically compare different examples. So I guess, how could we start doing this? Well, a common way to look at interactions, as I mentioned before, is by placing them somewhere along this cooperation competition axis. But of course, as we already noticed, interactions are really diverse and can depend on a bunch of different factors. And they encode much more information than just whether or not they were cooperative or competitive. So what I did early on in my PhD was this sort of thought experiment where we asked, can we come up with some kind of standard way to represent interactions while still embracing all of these attributes? And what we came up with was a simple list of nine basic attributes that we could use to describe interactions. So these ranged from whether or not the interaction depended on a particular spatial arrangement of microbes, for example, to whether the compounds involved were something like peptides to the habitat in which that interaction was observed. So what we could now do with this is represent any particular interaction numerically, according to whether or not it displayed a certain attribute. So with this framework in mind, what we did was we compiled a list of interactions that we sourced from the literature and manually classified them according to all these different attributes. So in the end, we can think of each interaction sort of as an entry in a catalog, which has its own sort of numerical barcode, if you will, defined by the presence or absence of these attributes. And as simple as this sounds, what this let us start doing was starting to perform these basic analyses to look now at how often certain attributes or interaction types were observed and how similar or dissimilar different interactions were with some surprising results there. And what we really look forward to with this is that kind of as new high throughput data sets continue to get generated, they can be compiled in some similar way, or it can be easier to compile them in this way for us to tackle these much bigger ecological questions. So with this, I would point out that this is by no means a new idea in general. So standardized formats have been designed for metabolic networks such as Keg, if you're familiar with that, as well as for protein sequences or protein interaction networks and 16S sequences. So we hope that by sort of applying these ideas to this world of microbial interactions, we can start doing the kinds of analyses and predictive studies that these other methods enabled. So I very much invite you to sort of look at our perspective piece here for more details on what we did. But with this, I'd like to transition over to the second part of my talk, which will look much more deeply into one of these interaction attributes, which is the question of metabolic cost. So I'll get to what that means in a second, but just to give a little bit of background, I would start by saying that there's this tenant in classical ecology called the competitive exclusion principle, which basically says that two populations that completely compete for a certain niche cannot coexist. So what this means is that if you have two microbes and both are competing for the same food source, the one that is less efficient at consuming it will die out. It'll be outcompeted by the other one. The problem is that this leads to a concept called the paradox of the plankton, which asks, how then is it that we see so many organisms coexisting in nature when there seem to be so few resource niches? And there's many strategies that can help us address this question, but the one we were interested in was that of metabolic cooperation or crossfeeding. So this happens when one organism produces a molecule that feeds the other, and that allows both of them to survive. The problem with crossfeeding is that oftentimes for the organism to produce that molecule, it needs to devote away some of its own resources and sacrifice its own biomass production. So this then can lead to the emergence of what we call cheaters in a community who only receive these benefits without putting anything back, and it raises kind of really big questions as to how this selfless cooperation can emerge in the first place. So there's been a lot of studies that have looked into the costs and benefits of this kind of crossfeeding, but we wanted to ask, what would happen if we sort of took this question of cost completely out of the equation? So essentially, can we look at molecules that can be secreted without an organism sacrificing its own growth? Which is a pretty big question, which kind of raises the question of how we would even measure this. So what we decided to do was use a computational technique called flux balance analysis, or FBA. And to use FBA, what we do is we get a model of a particular organism. So this is done by taking the genes and metabolic pathways that we know are part of a particular species. And we use them to build what's essentially a mathematical representation of all of the metabolic capabilities of an organism. So with this model, we then apply flux balance analysis, or FBA, which is a mathematical optimization technique that pretty much asks the following. So it says, given a particular organism's metabolic capabilities and given a particular environment, how fast is that organism going to grow? So at what rate will it produce biomass? And importantly to us, which metabolites is it going to produce, and does producing them impact the growth of the organism? So what's nice about this method is that we can ask these questions very precisely for any organism for which we have this kind of model. So now that we have a way to know when a particular metabolite is secreted in this quote unquote costless way, we can ask, okay, which molecule specifically can be secreted like this? We know that organisms produce metabolic byproducts all the time, like ethanol as a byproduct of sugar fermentation, for example. But how far does this really extend across a large number of microbial species and environments? And moreover, are these molecules enough to allow other organisms to grow? Or are they essentially just waste products, right? If they do allow other organisms to grow, what interactions could we see emerging from these? And importantly, how stable are these interactions over time? So the way we went about answering these questions is by doing a large set of computational simulations. So in each one of these simulations, we took two organisms, provided them with a growth medium, which contained two different carbon sources of two different nutrients. And we can also choose whether or not to provide oxygen to the organisms, which becomes important down the line. With this in mind, we apply flux balance analysis and see, first of all, whether or not the organisms grew. And if they did, which metabolites did they secrete in this costless way into the environment? And what we can then do is add these molecules back into the medium and feed it back to the organisms. So in this way, we're essentially carrying out an in silico co-culture experiment, where the organisms can feed off of each other's metabolic byproducts. And the really cool thing about doing this with FBA is that since we, in this study, did it using a list of 24 organisms, 108 different carbon sources with and without oxygen, we were able to sort of examine the impact of these costless metabolites in over 2 million simulated experimental conditions, which I think for anyone would be almost impossible to do experimentally, at least with current techniques. So with this method, now we can go back to our question, which types of molecules can be secreted in this quote unquote costless way? So what we immediately noticed was that, as I hinted at before, the presence of oxygen really was one of the biggest modulators of this. So we go into much greater detail on this in paper, which I'll mention in a bit, but as an overview, we do see kind of a wide diversity of molecules ranging from organic acids, as we expected, but also nucleic acids and peptides and even some more complex molecules. So now that we know kind of what types of molecules can be secreted, we can then ask our next question, is the presence of these molecules enough to allow other organisms to grow? And here, looking at kind of the core data set that we generated, we can say that many times it is. So what these bars are is they represent the total number of simulations that were performed. And because these media conditions that we provided only contain two carbon sources, the majority of the cases fell into this light blue section, which means that either only one organism grew or neither of the organisms grew. But what's really interesting, what we want to point out is this red set of simulations in the middle, which tells us that at the start of our simulation, only one organism grew, but the metabolites that it secreted were enough to allow the other organism to grow as well. So this sort of enhanced growth potential represented by the red region represents a 73% increase in the growth supporting environments with oxygen and an 83% increase in those without oxygen. So that was exciting just to see that these secretions can kind of lead to the growth of other organisms. And what's nice is that we can look at these cases a little bit more deeply on an organism by organism basis to get kind of a detailed picture of which organisms are the ones that are enabling the growth of others. And what we found were these sort of rich networks of these obligate interspecies interactions that show how, for example, this methylobacterium organism depends strongly on E. coli or how in a lot of these interactions, the salmonella organism is the one providing nutrients to all the others. And what I'd like to note here is that these organisms that we picked don't normally appear together in nature, but what this does provide us with is sort of a jumping off point to study any one of these particular symbiosis experimentally. And I'd point out as well that although these were generated with FBA, we were able to corroborate a number of these by comparing them to specific examples from our lab and also from the literature. So kind of building upon these interspecies interactions, what our dataset let us do was kind of zoom out and get an idea of the different types of interactions that emerge from these costless secretions. So we saw very clearly that even though organisms can be competing for primary nutrients, they can also be cooperating on the level of these metabolic byproducts. So we could see how often we got certain types of interactions like commensal unidirectional ones or these neutralistic bidirectional ones. And very importantly as well, there wasn't just one type of commensalism or mutualism. So what our dataset that let us do then is zoom in a little more and look at the exact interaction topologies that emerged. So we did this using a network motif approach which you can see represented here where the big yellow circles in each one of these motifs is an organism. The black dots are carbon sources and the purple dots are any metabolites that might be exchanged between the two. And these motifs ranged from really simple ones like non-interaction to unidirectional commensal ones which we abbreviated with C to more complex mutualistic bidirectional ones that we abbreviated with M. And what these motifs let us do is design dynamical equations to predict how the organisms would grow over time under these conditions. And I'd say this is a very important distinction since what FBA does is essentially just give you a snapshot of what the organisms are doing at one point in time. Whereas here now we could really ask how stable these different interaction types could be over longer periods of time. So just to go over this very briefly we made differential equation models for each one of these interaction topologies and parametrized them using the FBA data that we got and we let the organisms grow in a sort of simulated continuous culture environment. So what we asked was if both organisms survived at the end of these simulations I guess that would let us know a reasonable idea of whether or not these specific motifs were stable under the conditions that we set. So this is kind of a broad representation of the results which I'll take you through. So what we did in these simulations was we modulated the growth rates of organism one on the x-axis and organism two on the y-axis. So the areas in red within these areas in black that you see are showing the combinations of parameters that allow for coexistence of the two organisms. So by doing that we can see that for example this simple interaction n1a there's no combination of growth rates that let's both organisms survive which is trivial since obviously here organism two isn't receiving anything at all so it doesn't have any food it's not going to grow. The next motif is this sort of classic competitive exclusion where the species will only coexist if the growth rates of the two organisms are pretty much identical. So again that makes sense but then if we start having more complex interactions we start seeing pretty interesting things. So for instance this mutualistic motif n1b even though the organisms are competing for a primary resource the metabolic exchange is enough to sort of sustain them for a pretty wide parameter space. So what this study really lets us do is say okay well we have these costless metabolites that we looked at in this huge sampling of organisms and environments and then these metabolites can spontaneously lead to the emergence of certain interactions that allow other organisms to grow and it also let us have an idea of how stable these different interactions could be and as I kind of alluded to before in doing this analysis we've produced a really big data set of kind of very specific metabolically driven symbioses which we can follow up on on a case by case basis and this in itself can be really useful to this idea of synthetic ecology that I mentioned at the beginning which kind of if we have a better idea of how microbes interact in this way we can leverage that knowledge to engineer new interactions for a number of applications. So at this point now we've looked at how one of these interaction attributes that we identified at the beginning this idea of metabolic cost can shape microbial ecosystems. So I'll continue with the next section which looks at another attribute in more detail and here we're going to look at environmental composition and how specifically environmental complexity impacts the structure and function of these microbial communities. So what does this mean? Well one of the findings from our last study which was really important was that the composition of the environment so literally the nutrients that the organisms have access to was actually more important than the identity of the organisms themselves in determining which molecules would be secreted. And given how consequential we determined the secretions and these interactions to be it sort of adds to this body of evidence saying that molecular composition of the environment has a really important impact on what a community will look like. But there's a problem in sort of trying to understand this question of environmental composition since natural environments can contain thousands of different nutrients. So the result of this is that we sort of see a lot of studies of communities in very simple environments where we can get a lot of deep insight into the biology as well as a lot of studies that take natural samples but oftentimes just tell us the species composition. So bridging this gap is definitely really important. And we do know from isolated or from specific studies that the complexity of an environment is very important in a general sense. So there have been studies that have shown how for example the gut microbiomes of people under simpler western diets are really different to those under kind of more complex hunter-gatherer diets. And this concept is also important in lakes and oceans where an overabundance of one particular nutrient can cause these algal blooms that can destroy entire ecosystems. So what we want to know here is how do microbial community properties scale with increasing environmental complexity. So with a more diverse pool of nutrients. And right now I'll tell you about the portion of our study that looked into the impact on this environmental complexity on taxonomic diversity. So we focused on studying this systematically in a well-controlled experimental system in the wet lab so that we could really focus specifically on the effects of sort of a different number of nutrients. So what we did was we started with a number of bacterial organisms and we cultured them all separately overnight. And after that we combined them all in equal proportions and added them to these culture plates which contain a minimal medium and an assortment of different carbon sources. So either single carbon sources or more complex combinations of them. And we varied the number of different carbon sources but we always kept the concentration the same. So what we did was we then grew these communities in all these conditions and transferred them to fresh medium conditions so the communities would reach steady state. And depending on how complex the community was to begin with we measured what the taxonomic compositions looked like either using agriflating or 16s sequencing. So in total this experiment allowed us to look at the effects of this environmental complexity on over 280 different communities. And right now I'll go into detail on the results of one particular experiment where we tested a community that contained 13 species in combinations of up to 32 carbon sources that we assayed using 16s sequencing. But before I get into the results let's just look kind of very briefly at what we would maybe expect to see. So if we're basing ourselves again on classical ecological theory so specifically niche partitioning and competitive exclusion we might expect that the more complex an environment is the more taxonomically diverse a community grown in it might be. But we know now from some new experiments on computational work that this might not always be true. So for example we have evidence that when two nutrients are combined the community can exhibit sort of a non-linear growth response and we know from work done by a former grad student in our lab that even single carbon source environments can sustain very very biodiverse communities. And lastly we have evidence in an animal model that high nutrient diversity might not even be associated with high microbiome biodiversity. So this relationship between environmental complexity and community diversity is definitely more complex than I thought at the beginning of this project. So with that kind of background let's come back to our experiment and see what we get. So these are the relative abundances of our communities under each condition we tested. So the different environments are here on the x-axis and the bars represent kind of the breakdown of the different organisms that were observed at the end of these experiments in those different environments. And generally as we move to the right the environments become more complex. So here are the communities grown in single carbon sources. Here are the ones grown on pairs, on combinations of four carbon sources, eight, sixteen, and then the one community grown on all 32 carbon sources. So right off the bat we can see that there are varying degrees of taxonomic diversity. But what we see is that a lot of these communities are just dominated by a couple of organisms. And in fact none of the communities here retained all of the original 13 organisms that we put in. So to find out what happened we can start by only looking at the communities grown on the single carbon sources. So what we could do is kind of take these taxonomic distributions and compare them to how the organisms grew on their own, so not in a community, on these same nutrients. And by doing this we can see for example that this organism, acenidobacter baleii here, was able to grow on glutamine by itself. And it was also observed in the community grown on glutamine. So that's what these navy blue squares are. Alternatively you can see that this coring bacterium glutamicum couldn't metabolize glutamine on its own. That's on the top left. And it also didn't show up in the community. But interestingly we see that this space is dominated by sort of this light blue, which means that although an organism was observed to use a particular nutrient on its own, when you put it in a community setting it disappears. So what this indicates is that interest species competition, so competition for these resources and not just environmental selection is a major factor in these losses in taxonomic diversity that we saw. Interestingly though I'd point out that there were a few cases where the opposite thing happened. So for example by itself the pseudomonasputida in our experiments wasn't able to metabolize ribose, but it did show up when we placed it in a community context. So this could be an example of one of the cooperative cross-feeding interactions that we were really excited to see. So going back to our full results now we can look at what happens to the diversity of these communities as we increase the complexity of the environment. So we can look specifically at the species richness, which is defined as just the number of different species that we observed. And we see for example that in the 32 nutrient case we get four unique bacterial species. What was surprising initially was that there were even some single nutrient cases where we saw up to six different species which was the most out of any condition. Now as a different metric of diversity we can look at this quantity called the Shannon entropy which sort of defines how taxonomically balanced a community is. And here as well we see that some single and even some double and quadruple nutrient conditions are more taxonomically balanced than even our most complex environment. So as a result neither of these relationships showed a statistically significant increase with environmental complexity. So we sort of found ourselves asking ourselves what could be going on. Well one of the things we noticed was that a couple of organisms showed up consistently in our communities and when we look back at how they grew in monocultures so by themselves we can see that they had some of the broadest nutrient utilization capabilities. So in essence they were resource generalists. They could use a lot of different things. So maybe we thought it's something to do with who's a generalist and who's a specialist that determines the types of organisms that persist in a community and not just the number of the raw number of nutrients that we put in. So to quantify this what we did was we designed a dynamical model called the consumer resource model which predicts the abundances of different organisms based on how well they can use a particular nutrient. So essentially it gives us a way to define who is more of a generalist, who is more of a specialist, and then what the communities will look like with different balances of the two. And what's nice as well is it looks let's look at what types of metabolic interactions can go on. So we parametrized this model using our experiments as well as some flux balance predictions for metabolic exchange and we used this to make two models. So model A in which all of the organisms have the same chance of consuming certain nutrients and consumer resource model B in which you have a breakdown of generalists and specialists that are similar to the communities we tested experimentally. And what's nice about the model is that we can simulate these many many times with randomly assigned nutrient utilization probabilities so we're not just limited to the specific experimental conditions we tested. So here again in a different format are our species richness and Shannon entropy values from the experiment and let's compare them to the first modeling result. So this model tells us we should expect higher diversity than what we actually saw. And again here you only have specialists so the likelihood that the that the organisms might overlap in terms of niches is lower. But now looking at what happens when we have a mix we see a much stronger agreements with our experimental results in which we see much lower levels of taxonomic diversity. So this was striking and it was really nice to see since these models were pretty much parametrized based on nutrient utilization capabilities and were able to kind of very nicely reflect what we saw experimentally. So overall what do these results mean? Well it suggests that kind of contrary to this classical intuition environmental complexity is not a requirement for maintenance of taxonomic diversity in microbial communities. And we know that now by having done this sort of systematic study of how the same community responds to increasingly complex environments. This gave us an idea of an idea of a strong role that competition plays in dampening the diversity of these communities with some instances of metabolic cooperation. And we found that what appears to be most important in determining the composition is of course the identities of the nutrients but also the distribution of generalists and specialist organisms within these communities. And another important thing that we gained again just like the last study is this rich data set that correlates specific environmental compositions to community compositions and also their their growth phenotypes which I couldn't get into today. And I think we can use data sets like these as jumping off points to sort of design complex communities again by modulating their environments. So there's a lot more that we did and I invite you and to look at our preprint up on the archive for this. This article is actually now impressed so hopefully you should see it out very very soon. But very quickly I want to move to our last section where I'll briefly talk about some work that now leverages what we've learned in order to design new synthetic communities. So what for me has really come out of this work is sort of a deep appreciation for how even small changes in environmental composition can give you very very different community structures. So part of that is really scary but part of that is exciting since it sort of gives us a chance to ask the reverse. So if I want a community with a certain structure or certain behavior is there an environmental composition that can get me there. And this idea is really attractive since modifying the environment is often a lot easier to do than modifying the genes inside the organisms themselves. So what this puts forward is the possibility of sort of rational environmental modification as another design tool for this field of synthetic ecology. The problem as you might have guessed is that there are just so many possible combinations of environmental nutrients. So for example if you wanted to test all combinations of 20 nutrients which is an amount that's far lower than what we would expect in nature you would need to do around 1.05 million experiments so not very accessible. And even if you wanted to rely solely on modeling just slightly increasing the complexity of these environments would very quickly get you to quantities that are computationally intractable. So we need a way to search the space of possible nutrient combinations to get us to a specific community composition. And there's a lot of methods that can help us do this but the one we ended up choosing was a simple genetic algorithm. So what this is for those of you unfamiliar is just a search process based on natural selection which works as follows in our case. So first we randomly initialize different environments with different nutrients. We grow our communities on them and see how they perform. So these are just sample growth curves. And say here we wanted the most taxonomically balanced community so we could rank the environments based on how close they got us to that goal and we would pull the top ones to produce a new generation of environmental conditions based off of them. We can also allow for some random mutations to happen so those are the little red squares that you can see in order to fully explore this combinatorial space. So we can then take these environmental conditions and test them again refining them over and over over the course of several generations so what should happen is that over time the process should converge to an optimal set of environmental conditions for the goal that you had in mind. So now what we can do is take our community and our environments, grow them up, measure how the community did in the context of the objective that we want to optimize and run it through this algorithm to help us select the new set of environmental conditions. What we did here was since we don't have the experimental data set to be able to do this fully we relied on simulated community data again relying on flux balance analysis. So very very quickly just to kind of end I'll show you some of the things that we can do with this. So let's say we have a multi-species community and we want to make it as taxonomically diverse as possible so we can set again the Shannon entropy as the objective and have the algorithm find us environments that maximize it. So here we can see that for this community the maximum Shannon entropy is around 2.5 and the algorithm is able to get us there within just 20 generations or iterations of this cycle and if we wanted to have a community that was dominated by a particular organism we could do that as well and if we wanted to maximize the number of metabolic exchanges it can find environments to do that and we tested a bunch of different objectives in addition to these. And what's powerful about pairing this with flux balance modeling is that for any of these solutions we can go more deeply into the metabolic networks to see which interactions and exchange patterns are going on that can lead to these particular community phenotypes. So I realized that I've thrown a lot at you so I'll stop here. So just to review this part what we've been able to do is combine sort of a search algorithm with flux balance modeling that gives us environmental compositions that are likely to give us community structures that we want. Now of course here we tested it on computational data which has its limitations so first and foremost is the question of accuracy so even the best genome scale models are bound to have prediction errors and even though in this study we only used manually curated high quality models we can't be sure that they'll necessarily give us the right answer so this is why we designed this framework with the intent of integrating with an experimental system. So a nice thing about the genetic algorithm is that you can test a reasonable amount of experimental conditions at a time so we envision a setup whereby you do an experiment, feed the data into the algorithm and it suggests new compositions to try. And with the advent of new sort of ultra high throughput technologies you could imagine sort of the sci-fi future where this is all automated and you get your target community through this kind of combined computational experimental process. So just to review what my work really focused on and what I talked about today was first looking at how complex microbial interactions could be and starting by kind of proposing a framework that embraces all of this complexity all this heterogeneity to enable ways to quantitatively analyze and compare these interactions across different studies. We looked at how this question of metabolic cost one of the attributes that we defined in our review article could define what a microbial interaction could look like and from here noticing that sort of the composition of the environment was a crucial determinant of these interactions we looked specifically at sort of this under explained question of how the complexity of the environment itself contributes to community ecology and then based on what we learned here we designed an evolutionary algorithm that can let us design new microbial communities with specific structures just by modifying the environment. So in sum what I what I hope this work does is is shed light on the mechanisms that underlie microbial interactions with a specific interaction focus on how the environment affects them. And what I find really exciting is that it gets us closer to being able to sort of rationally design communities via environmental modification and definitely hopefully shows how a combination of experiments and modeling can help us eventually achieve these these promises that I mentioned of synthetic ecology. So with that just take a quick moment to thank the people who supported this work. So a lot of this or all of this rather was was done under the supervision of my advisor Daniel Segre and I'd also like to thank my dissertation committee and members of the lab both past and present for for all of their support. Also thanks to my my collaborators and people who give me advice through all of this as well as obviously the bioinformatics program at BU and for the funding sources that made this work possible and before ending I would just like to give a shameless plug for another super cool funding source which is the James S. McDonald postdoc fellowship in understanding dynamic and multi-scale systems. So they are currently funding my postdoc research and the application for the 2021 cohort just opened up pretty recently. So I have to say this is an awesome funding opportunity for anyone looking to do research after their PhD not just in the area of biology but in all areas of complexity research. So I would say check them out at their Twitter handle or feel free to reach out to me directly if you have questions about the fellowship or the application process. So with that thank you for for taking the time to listen. Thank you for the invitation again to join me today and I'll stop now for questions. Wow really fantastic talk. Thank you Alan. At this point we will end the live stream and take questions in another video chat with RSVPs for this event which have received a link to kind of hang out with you after and ask an endless number of questions.