15:05:28 Okay.
15:05:31 Hi everyone I'm Daniel Segre.
15:05:33 I'm going to give you a little bit of an overview of metabolism and this is obviously a very wide area. I'm going to focus on some specific aspects of metabolism in particular metabolic modeling that it's relevant for microbial communities.
15:05:50 And I should say this is, I'm going to talk for a little bit about our own research, mostly really my perspective and there is a lot of subjective aspects of this that the choice and the way I chose to go through this.
15:06:04 I'm happy to you know go through faster if things are obvious to everyone.
15:06:09 But obviously also linger on things that need more. So, why I want to start by saying that, why do we want to study micro community metabolism I think it's obvious to all of us, as, as we heard yesterday from auto.
15:06:22 There are key determinant of community structure and dynamics, so that that's pretty obvious but I want to highlight also there are a lot of practical applications of why specifically, we want to look at the metabolism and microbial communities from metabolic
15:06:37 engineering efforts to produce biofuels or other compounds to obviously the human microbiome side.
15:06:45 Understanding how microbes that live with plants can be make better crops. And also, I think, very importantly, the role of microbes in climate changes we heard that, as you know, Mike was can change the faith of a planet.
15:06:59 So I think there is more and more interesting trainers then whether we can understand that and potentially leverage leverage them.
15:07:06 Another aspect that I find fascinating and you can ask Josh For more on this but I've been interested in the origin of life for a long time. And when I look at the map of the call the whole metabolic, the collection of all metabolic reactions across all
15:07:26 living systems. It's clear that there is some signature of the early stages of life on our planet in this map, and in one of the questions is, in particular, whether by knowing what which of these compounds might have been present on early Earth, kind
15:07:40 of solve this inverse.
15:07:45 Connect the doc problem where we actually know the final picture. We know many of the dots may know which lines were there at the beginning, but we don't know the order we don't know how this arose.
15:07:54 Another aspect which is, we always fascinating even, you know after seeing this and I'm sure most of you are familiar with this. This is ATP synthesis.
15:08:05 Big multi protein complex that is responsible for the key stage of respiration, in
15:08:14 many living cells, and I finally just amazing and one of the reasons finding amazing. In addition to just the beauty of this is that this is really a process where you know when you talk about Redux chemistry and what electrons, doing in biological systems.
15:08:29 What is what I find just amazing about this is that the electrons and protons that are generated from NADH through the TC cycle really go through different ways in this process the electrons go along the membrane.
15:08:45 And through the different steps of the electron transport chain and pump and pump the protons out that can go through back and produce ATP ATP synthesis so there is discharged separation and electrons really behave a little bit like electrons in a wire.
15:09:02 So, more on this perhaps later but this is just a topic that I find absolutely fascinating and I'd love to talk more about with all of you.
15:09:12 Now, I will spend a few minutes talking about how to navigate metabolism and again, I apologize for those of you that have more biochemical background that I do pretty much zero.
15:09:23 Will. Will I hope this is not going to go through this fast but I want to show how I personally think of metabolism and navigate through this complexity, trying to make sense of it.
15:09:34 So, one thing that I find particularly useful when I look at pathways, this is the DCA cycle is to look at the what I like to think of as the arithmetic backbone of metabolites so if you look at this network, this is, You know the cycle, this is quite
15:09:51 complex and, you know, you can try and memorize all the details or understand there is a lot of interesting aspect to every single reactions, but you can also try to simplify this one very extreme stage where you can just look at the number of carbons
15:10:07 in this network. Okay. And when you just look at the number of carbons, this is what emerges right you have basically pairs, two carbons that come in positive way, they're fed into this cycle.
15:10:23 And you can see that the arithmetic simplicity of this network. And you can do this spiritually for any metabolic pathways just visualizing, think of the metabolic backbone of every single element so this is the carbon back one of course, and it doesn't
15:10:33 capture a lot of the details that are relevant for this but it captures some of it which I think is actually quite important and interesting.
15:10:42 And in this case, the TC cycle running this way and produces co2 produces the NADH that is driving the electron flow that I was mentioning earlier.
15:10:55 But there is also reverse this a cycle that can run backwards and you can see from this backbone, that when the reverse this is this a circle run backwards, you can actually fix carbon to produce carbon carbon bond in this is in fact one of the hypothetical
15:11:11 keep pathways relevant for early stages of of life.
15:11:16 If anybody's interested in this by the way we've built a toolbox for playing with artificial chemistry, chemistry, we're all you have our backbones of letters and so it's a fun way to think, you know, to play with artificial chemistry.
15:11:32 Yes.
15:11:34 Oh, microphone.
15:11:41 Sorry, all that for a very simple question.
15:11:45 Is there a simple understandable reason why it has to be cycles, like kind of the naive thing when you first thing was, how was like oh you take this thing and break it down it's a line.
15:11:54 And somebody told me that, like the CCA cycle, there's another hypothesis was actually align and until it became close or something.
15:12:06 There are in fact in some organ is have only half of that is a cycle.
15:12:11 So this is still existing.
15:12:15 The reason you may have cycles in multiple instances is their auto catalytic nature, and one concert there. In fact, many more than, you know, than the classical ones possible auto content cycles in metabolism.
15:12:30 So I think my, my first answer to this is really that the auto catalytic nature of the cycles is crucial component but I, you know, I'm happy to think about this more together and.
15:12:48 Yes. Ancient systems, is it also imply that they have to be communities that this also single organism.
15:12:59 Thank you, thank you, thank you for the question. So, we know one thing I forgot to mention here in fact is that right when you think about earning the top here when you think about early metabolism.
15:13:08 One thought that I find interesting is that life has always been an ecosystem. It was an ecosystem of molecules there has never been in color in isolation.
15:13:18 So, one of the interesting aspects of studying the early history of metabolism by looking at the whole metabolic chart is that they're not constrained to reactions that are together in individual organisms today, and you can look at potential couplings
15:13:32 that may have existed in, you know, previously and may not exist anymore. So, I think, yes, given the extensive horizontal gene transfer, and there's probably been a lot more shuffling done and do today.
15:13:52 So, staying now for a second with this simple backbone. This is visualizing something that I've been brought up yesterday by auto and you're probably very familiar with the difference between fermentation and respiration.
15:14:05 In this case, shown here is the equivalent of glucose, going through fermentation where there is some unbroken carbon carbon to that could be secreted or the carbon can be fed into the TC cycle to produce ATP through respiration.
15:14:25 And one thing I want to point out is that this is very relevant for ecological interactions in one of this thing you can see here is the yield of ATP risk of fermentation.
15:14:37 Typically, gives a net production of ATP per glucose, being broken down, whereas respiration give 32 or more ATP so it's a huge difference in the yield of ATP per unit glucose being broken down.
15:14:53 And this is obviously a very interesting question in terms of ecosystems there is this potential rate yield a trade off. And that's something that also I think it would be an interesting topic for discussion there.
15:15:06 To my knowledge, there are different possible explanations for this Terry's work is obviously a very important one in this context of the cost of protein production of mentioned this later again.
15:15:19 But there is also indications of, you know, which I'd love to hear more thoughts from those of you that are familiar with non acclaim or thermodynamics, but there are hypotheses about the very, you know, the rate of these reactions and the rate of these
15:15:34 two pathways being substantially different for for basic thermodynamic reasons.
15:15:40 So I think that would be another interesting topic for discussion, but it's it's pretty clear that, you know, organisms that choose this strategy as opposed to this may have, again, either some protein cost reasons or some other choices that should be
15:15:57 very important or to give up this additional experience.
15:16:01 The other parts of this which may explain in some cases some of these choices, is the fact that some of the byproducts that could be produced upon fermentation, and be used as carbon sources by other organisms.
15:16:14 And if there is where they're suppressing the of these interactions, for example, one organ is providing carbon to another organs that can provide fix nitrogen, you could imagine this being a useful strategy.
15:16:25 Yes.
15:16:33 Can Can these pathways peace superimposed with each other. In other words, can, can you have a mixed strategy or Yes, believe the answer is yes.
15:16:44 And
15:16:47 there are I mean mixed fermentation, there is, I think combination of respiration fermentation. I think this is still active area of research right what happens, for example, when organised have the availability of both some, you know, broken down compounds
15:17:02 acetate and glucose can they use both, there is a lot of really interesting recent work on this.
15:17:08 Yes.
15:17:16 Yes. Yes, thank you. Sorry I obviously I forgot to mention this, this depends on oxygen or an electron acceptor availability and maybe just going quickly to this just to complete the picture right what happens to these electrons.
15:17:31 Ultimately, they Rican, re, re, join the protons and with oxygen as the electron acceptor they form water. So, in order to for this process of respiration to take place you need the oxygen.
15:17:49 So again, you know, I think this is useful I personally find it really nice to look at this backbones, but I just want to remind that this is, you know, highly oversimplified.
15:18:00 It's still I think useful to look at this different sub structures in a modular way you can look at the nitrogen backbone or other backbones. But then there are other things you obviously want to add on top when you think about how metabolism works.
15:18:13 The structure of electron donor So, except there's the thermodynamics the kinetics component compartmentalization and what I'm going to talk about some of this next.
15:18:23 So, the, the thing that makes her metabolism really complicated and interesting. In the end, in my mind is that it's really a multifunctional resource allocation network, and you know this very same pathway that we just showed that you know we can think
15:18:40 of the basic energetic engine of the cell that produces ATP from glucose is also responsible for producing many of the building blocks that foreign protein so you can see here, from this really nice paper, how the different units in stem from different
15:19:01 locations along this network.
15:19:04 And what this means is that as an organism produces its own ATP for for energetic reasons has also to siphon out enough amount of each of these amino acids to build the proteins that form the same amino acids, the same sort of the amino acids that will
15:19:20 form the proteins that will catalyze those reactions. And it's not just a random amount of course they have to produce the right amount of each of these amino acids.
15:19:29 So it's really a, you know, even just if you look at this very tiny portion of metabolism, this is already a very complex resource allocation problem in my class I usually draw much much simplified version of this with just two amino acids in ATP production
15:19:43 and I challenged my students to try and find a solution that steady state solution of this network it's pretty much impossible already for a very small network you try to balance everything have the production of different amino acids in a precise way,
15:19:57 it's very hard to balance this manually. And the reason we need to understand this of course is that each organ has has to produce its own biomass is to produce more of all the amino acids forceful if it's nucleotides and so on.
15:20:10 in the right proportions to make a new cell.
15:20:13 This is a diagram of the composition of an ecology cell that is used for the color flux balance model in an earlier models from Paulson and colleagues.
15:20:24 What is interesting is this state is typically very difficult to obtain. So for a long time, the spammers composition was taken from the E. coli Bible by 900.
15:20:35 And it's still hard to get new by masking position from New Orleans and I think this is one of the bottlenecks for making good models for New Orleans, we can talk more about this there.
15:20:47 Yes.
15:20:50 Microphone.
15:20:54 What if you overproduce it an amino acid, you can still excrete it to the for everybody else to us so it may be this balance in it's not as difficult as it seems.
15:21:04 Yes, so that's a great point. And in fact, I'm convinced and you know there is evidence if you do, for example flux balance models under many different conditions, you'll find that there are a lot of, you know, opportunities for secretion of amino acids
15:21:19 are the byproducts that can in fact used by other organisms. But still, I think the, you know, a very small imbalance could create a very large pool in a short time.
15:21:29 So I think, by and large, even if there is production have some compounds.
15:21:35 The overall you need to be very close to balance but it's an interesting question. I don't know if anybody has assess quantitatively how far can you be from balance before just even over producing wouldn't be sufficient.
15:21:47 Good question.
15:21:49 Okay. So, we have this biomass composition and the next question is, how do we use what we know about the genomes of Oregon is to build quantitative models of metabolism.
15:22:03 So, the, the first step in trying to make a model for a whole organism is to try and reduce the set of reactions from this complete network graph, or all organisms.
15:22:17 Through the genome and from literature searches typically for an individual organised to try and build an organism specific network. This is the equivalent network from eco site.
15:22:26 In a typical microbial networks, as far as we know today has of the order of 1500 reactions about 1000 metabolites. This varies, of course from organism to organism but that's just the order of magnitude.
15:22:40 And this is a complex process, as you can imagine there are a lot of issues and again, I'll get back to this but the notation of genomes is a big bottleneck in there are many different methods in a lot of models for model organisms such as the call a
15:22:56 nice has been built throughout many years for a lot of manual curation so we have pretty good models for those organisms. But now people are interested more and more in making ultimate.
15:23:18 I find that will construct models from genomes. I want to point out the DOE department energy k base has a set of tools for meal building models biocide. I think as engines for making models from genomes.
15:23:21 And there are other algorithms such as this carve me from Daniel Machado and colleagues. The try to fill the gaps if you find gaps in the networks and you know that Oregon should be able to produce an amino acids.
15:23:33 You can try and fill these gaps, but this is a whole art and a whole subfield in itself, which I'm not going to go into now, but I'm happy to discuss more later.
15:23:41 And this is of course important when we try to model complex communities, right.
15:23:46 So, one thing that is interesting is that from this network you can construct what is called the sticky metric matrix which enables a lot of subsequent mathematical analysis so let me spend a couple of minutes, describing for those of you that are not
15:24:00 familiar what the symmetric matrix is, and this is I think best illustrated by the time model like this, where we have a cell with three metabolites in a few reactions, notice that our interest our reactions in some transport reactions.
15:24:15 And you could imagine having a flux, a rate of each reaction that is somehow a function of concentrations and so on. With Nicholas mental equation, but you can represent in, try to write the differential equation for each metabolite as a function of the
15:24:29 incoming and outgoing flexes. So for example, or metabolite a you have outgoing flux be one outgoing flux be to be three is producing it and it's coming in through the exchange flux be one so you'll have this equation you can write this set of differential
15:24:45 equations and if you write it in matrix form, where you have a vector of concentrations in a better fox is the matrix that allows you to write this equation in a compact way.
15:24:56 Is this your sticky metric matrix of the network. Notice that you can relate in different ways, it's a very useful mathematical tool, because you can look by column for example by column, each column will be a reactions, this would be reaction one that
15:25:10 consumes metabolites a produces metabolite D, or you can look by metabolites so you'll see what reactions each multiple metallic participates in.
15:25:20 This is an example of a psychometric matrix for the beginning of like colleagues is what again you can recognize, for example, glucose, being imported into the cell.
15:25:30 This matrix get, of course, really huge but there's still much more tractable than trying to deal with these graphs. And I should say also sometimes people don't realize the metabolic metrics are bipolar type graphs you have notes that are reactions and
15:25:44 notes that are metabolites, you cannot represent them well as simple graphs because you will create weird shortcuts.
15:25:52 You know more on this if anybody's interested but essentially this is a representation of this by Partha graph. And it's very helpful for doing any follow up, mathematical analysis of metabolism.
15:26:04 Any question.
15:26:07 Okay, so now once we have this network and the second metric matrix, we're still not done right so this is just the representation of the architecture of a network.
15:26:19 But we wanted, what we want to obtain is an estimate of what a network can do the phenotypes metabolic phenotypes that are possible for this network, and in principle right what one would need to do is to write for each of these these, each of these reaction
15:26:32 rates that were shown before on the arrows, you can write the corresponding mechanics meant an equation that will tell you, oh that reaction right will depend on this V max that is proportional to the amount of the enzyme catalyzing that reaction.
15:26:45 And then there is the saturation function for the substrate. Except that, you know, this would be the simplest possible reaction for a typical metabolic reaction you'll have an expression like this you may have reversibility multiple substrates and so
15:26:58 on. A lot of static regulation. And you can imagine, of course, having equations like this implies having a lot of unknown parameters, a lot of non linearity is not necessarily an easy way to model a whole metabolic network of 1000 reactions.
15:27:12 So, it's still there, interesting, ongoing research direction that try to model the whole metabolic metric of a cell using dynamical equations like this.
15:27:23 Mark was covered for example is doing beautiful work in this area, but I'm going to switch to a world where this parameters don't matter so much, which is the old world of the geometry.
15:27:35 And this is what is really the idea of flux balance analysis, which was introduced early on. By Terry puppets artists I don't know hey, this is the reference, I think this is one of the first papers that use this idea of strict geometry and balancing
15:27:48 sticky symmetry. In order to understand the global functions of metabolic metrics, without dealing with the details of kinetic parameters.
15:27:56 And this was then developed further by Bernard Paulson and collaborators and now it's it's really, really expanded to a whole field.
15:28:04 And the idea is to really try to do a resource allocation resolve. Mathematically, this resource allocation problem I mentioned before we have new trends coming in yourself.
15:28:22 compounds that make biomass and there may be production of byproducts as an outcome. So, for those that are not familiar to go quickly through the basic assumptions of flux balance analysis, the first, as the name says is blocks balance.
15:28:31 So the idea is that we can assume a steady state for the system, which again implies that the net some of the flux is producing or consuming each metabolite is zero, which translates into simple linear equations that, that somehow make life much easier.
15:28:48 One thing I want to stress and point out and it's important to remember is that even if we from now on we'll deal with this linear equations This doesn't mean that these boxes are just, you know, we treat them as black boxes but they're still in principle
15:29:03 dependent on all the parameters that concentrations and so on.
15:29:06 They were by making this approximation we essentially say we give up, understanding the detail metabolic dynamics.
15:29:15 What happens, the metallic concentrations we focus on the flux is only as the main variables of our system.
15:29:22 And of course you have one such equation for each metabolite.
15:29:26 You can also write this as an SV equals zero equation or SS dystopian metric matrix so this is just one element one row of the second metric matrix multiplied by the column vector fluxes.
15:29:52 And this will give us certain constraints. So you have one constraint for each metabolite, you can add constraints that pertain to the specific conditions on which are running an experiment that you're trying to simulate, for example, you may know the
15:29:57 amount of glucose that is coming into the cell, either from measurements, or from calculations, through kinetics. And, and you may know for example that some reactions are effectively reversible.
15:30:06 And this is very important as we'll discuss soon, but somehow knowing that certain reactions under standard physiological conditions, and to happen only in one direction, help a lot constrain the problem of course this is never rigorously true in principle,
15:30:23 everything is reversible but this is a very useful practical step to make this model.
15:30:30 So all of these constraints together define what is called the feasible space, which I like to think of in the two dimensional projection as this context, Polly Hedren know each of these constraints right each of these.
15:30:46 Each of these represents a hyper plane of dimension dimension and minus one in the end dimensional space of flux is to have many hyper hyper spaces like this that intersect and then you cut half spaces by taking this inequality so you'll end up with this
15:31:02 on the total return and what is convenient about this is that if you look for extreme points on this.
15:31:09 In this one, you will have a discrete number of points to look for.
15:31:14 So what is being done often in flux palace analysis is to try and find the subtle flexes that maximize a given objective function for example, the sort of flux is that allow the cell to produce biomass in the most efficient possible way.
15:31:28 So this allows you to search engine within this physical space for the optimum the point that maximizes that objective function, and what is nice about this is that this can be solved with linear programming which is very efficient in a fraction of a
15:31:40 second you have a prediction of all the flux is in the cell, the growth rate of the cell uptick in secretion rates and so on so it's very mathematically convenient and simple.
15:31:52 Of course, you know we have to remember and we'll get back to all the approximations we made, but this is a useful way of obtaining predictions about the metabolic network might be able to do.
15:32:02 Microphone.
15:32:09 Yeah, I guess so last last slide just to make sure you understand the approximation. So is it really that last blue box that's the approximation that I should think of the other three is being satisfied in the kinetic world to all of these are approximation.
15:32:21 So, the steady state is clear an approximation and I think what is often not spelled out I think in flux balance analysis is that, so the steady state proclamation seems reasonable say for chemo chemo side or something like this.
15:32:34 But even a chemo start you know it's kind of it's going to be the average of many cells right over time and so on. And one has to be careful in some cases clearly not a good assumption when you and you know you start doing that, look at dynamical process
15:32:55 Yeah but, but, and then this is perhaps the most debated and approximation that has been made, but I should say, you know, I mentioned here sampling, you don't have to do use of maximization, there are many other ways to explore this network you can sample
15:32:58 So that is one approximation, this capacity constraints are also an approximation in the sense that we don't really know right unless you measure directly in there.
15:33:07 this space, try different objectives and so on.
15:33:20 Now the other thing I should say right which is even, you know, the bigger picture level that the other approximation you make here is that, somehow, the proteins are not you know the protein cost is not playing a role, and there is no you know regulation
15:33:34 is somehow.
15:33:36 Part of the, of the optimization process so by saying are maximizing growth rates. It's as if I'm saying well I don't really know how the cell does it, but there is some regulatory process that will.
15:33:49 So, this is a summary of some of the things that I think is important to remember, kind of what I think of the pros and cons of this approach.
15:33:57 So I think, you know, no matter whether it works or not having sticky metric matrices represented organism is a really good formal representation of what we know about metabolism.
15:34:08 And it's very fast as I said you can you know do tons of different explore different conditions perturbations and so on because it takes so little time to run the simulations as opposed to kinetic models.
15:34:19 You don't need to know the kinetic parameters again. You know, it's good.
15:34:25 And the other thing is that self care about Fluxus right though when you, when you think about ecosystems and secretion of compounds, what matters is the rate of secretion of compounds the rate of growth and, and we care often about flexes we do metabolic
15:34:39 engineering you want to know the rate of production specific compounds that are useful to you.
15:34:44 But of course I have to remember this other aspects so first of all we're limited by knowledge so if you have incomplete annotation, there isn't much you can do mean, there are things you can do, but it's a big problem.
15:34:56 The other thing that is really really important is that you know I get this question many times will tell me oh I have tons of makeup longest data can we collaborate and if it's Interstellar metallic concentrations there is nothing I can do with my models
15:35:07 to connect to that, you know, and that's not completely true. There may be something that can be done but it's it's complicated involves thermodynamics FBA more on this in a minute.
15:35:18 But it's not straightforward. The other thing is there is nothing that really.
15:35:24 Yes.
15:35:28 That's okay, sorry,
15:35:33 supposing that the psychometric matrix is right. I think all of these cons that you're describing. They all shrink the poly top right, like that none of them can change the shape and have it you know move outside the boundaries of the original pilot hope
15:35:47 is that is that is that right, assuming that this thing commentaries Yeah.
15:35:53 Yes, Yes, Yes, I think so that's that's some extent, there is this. Yes, kind of beauty of the fact that the constraints of the commentary is kind of absolute once you know that.
15:36:15 Yes.
15:36:15 Turn it off.
15:36:19 Okay.
15:36:19 This might be the same question but it seems like if we don't know anything about the metabolite concentrations. And so what's possible on that feasible space of fluxes.
15:36:28 They're actually combinations of fluxes that aren't aren't possible, because they might require some impossible combination of. Absolutely. Yeah, so there's some subset of the physical space that's actually feasible.
15:36:38 Yes, yes, absolutely. So that's yes that's another side of the same question so this is you know that politics is the maximum allowed space but it doesn't mean that there are regions in that space that are effective we are not allowed for more than one
15:36:51 reasons, we'll see another reasons.
15:36:56 When you apply FBA in practice how much sort of degeneracy Do you get Do you get a lot of things near the optimum that are just like totally different compositions of fluxes typically Yes, surprisingly somehow I you know I didn't expect this when we first
15:37:11 were doing this, but there are often multiple equivalent Optima so you know if you look at this. What this means right this the line that defines objective function will be parallel to one of the faces of this polynomial cone and.
15:37:24 And yes, you often get that there are ways around this, there is this parsimonious FBA idea where you after you maximize for growth rate for example you can minimize the sum of the absolute values of the flock so you can take the most parsimonious that
15:37:39 are fluxes in you know in practice when you do FBA you'll see sometimes that there is a metabolite that goes out, you know with the flax that is maximal outbox and then comes back, and you know, but but it's not you know there may be cases that are more
15:37:53 subtle and might be best.
15:37:57 Thank you.
15:38:03 Justin to go back to these question I think I understood your answer is saying yes, these constraints all restrict the poly dope, irrespective of the steady state assumption.
15:38:17 But is that really true.
15:38:19 So for example a constraint that's commonly used an FBA is like a crowding constraint right, a molecular crowding constraint.
15:38:26 I haven't thought this through but I'm just wondering if if there might be feasible metabolic flux is outside this this front outside the steady state condition.
15:38:39 Maybe it's not an easily.
15:38:45 I think I know we can discuss but I think the fact that right, it is basically mass conservation, right so you can't violate that I mean one interesting question if you start in people try to do this to make bigger flux balance models that include every
15:39:00 single day every single protein and so on.
15:39:05 You know I'd have to think about this I don't have an immediate answer whether so maybe this is something to revisit my intuition is that this is really an absolute maximum set of physical states, but I haven't thought about the protein.
15:39:17 Protein part than that module later.
15:39:22 Yes.
15:39:30 I don't think that microphone is on okay yeah superficial understanding of this But my question is that doesn't radio the trade off for out of this model, if you play with it right do you observe radio trade off as, as you mentioned at the beginning of
15:39:49 your talk. Yes, but it's complicated.
15:39:52 I actually I don't I did did we answer is that I don't know I haven't specifically try this.
15:39:58 What, what might happen right if you maximize growth rate ensure limited by oxygen.
15:40:03 What will happen is a typical flux balance model will will take all the oxygen that is possible and all the glucose ferment started respond whatever portion, you can require and then ferment the rest.
15:40:21 So I know the truth I don't know, I don't know, I guess it depends on under what constraints.
15:40:26 But I think that's something to, interesting.
15:40:32 Yeah yeah so that I agree, I agree that's that is not.
15:40:37 So I don't know, I think that's a good question.
15:40:38 I just want to follow up and set this question, I guess.
15:40:42 Because you reply that that's just mass conservation. If I understood correctly or maybe to follow.
15:40:49 Wouldn't I mean it's it's ignoring the possibility of something accumulating or something is accumulating that it's you know you haven't yeah yeah so if something is like day night cycle and during the day it accumulates and during the night it's used
15:40:57 up or glycogen in the liver or whatever that seems like yes yes yes so so that is under the city state or something. Yeah.
15:41:11 Yes, there is accounting for that. Yeah, yeah. So, but not accumulation.
15:41:23 Right. Yeah, right. You can Yes, yes, Yes, absolutely.
15:41:30 Yes, sure you had steady states, assuming it sort of the system which is a fixed point and fluxes right. could you.
15:41:40 Is it possible to simply extend this to, for instance, oscillating solutions in the flux space.
15:41:54 Yes.
15:41:57 That could that, not with this framework, but I'll get to a case where we can actually do this in a minute.
15:42:00 capacity so what's the question.
15:42:14 how do you choose Oh no, this can be any number of constraints. I mean, this could be any typically it's for each elemental source in the network you'll have typically a capacity constraint for the carbon coming into the network the nitrogen, phosphorus,
15:42:28 the sulfur and so on.
15:42:31 Very.
15:42:41 Don't pass the microphone.
15:42:49 One of the most unsatisfying things I find with FBA is that I'm just no way to track energy.
15:43:09 No, just because you don't know the, you don't know the conversion ratio. I mean, really only DHL how many electrons get kicked out. How many ATP is generated.
15:43:09 And, yeah, and also what all kinds of futile cycled that might be running right so yes, yes. Yeah, thanks for bringing that so I think typically those know the ratio ready, how many ATP is you can produce per proton that is being pumped out.
15:43:26 Those are empirical numbers that are put into the models, when whatever input.
15:43:35 I
15:43:43 know that multiple pathways can turn on that they use different ratios, from a practical standpoint, when we try our best to count every process that generates energy to consume energy and using different ratios.
15:43:51 We cannot account for at least 50% of the ATP.
15:43:55 And I think I've talked to many people. I mean, the two people that did this sort of thing.
15:44:04 They share our experience. Yeah.
15:44:04 No, thank you think I completely agree so the ratio is one there is the maintenance I don't know if that's what you're referring to, there is a lot of maintenance maintenance defined as a strong growth limit but when you're growing in glucose, faster,
15:44:18 and you cannot account for 50% of the energy and that's a bit of a problem.
15:44:23 Okay, I'm happy to hear, you know, don't have to hear more but
15:44:29 I mean energy in general is a problem I mean it comes back through thermodynamic up to some extent, but I don't think that solves the problem you're asking about it is generating appears to be generated much more energy, unless it's like just just converting
15:44:44 this ratio the way off but it's converting.
15:44:47 Very little of the NADH to ATP, which is also a kind of a futile cycle. yeah.
15:44:58 Yeah, I don't have an answer to that i think you know that's a good point and, you know, the there is a weakness and it's not should be in my list of cons, because.
15:45:17 So I want to mention briefly.
15:45:21 So I want to mention briefly somewhat, I mean, there is a lot of effort and it's, you know, as Tara was saying and as you may know, it's, it's not easy to test these models and the kind of key type of predictions that are testable are either direct measurements
15:45:35 of boxes with carbon 13 labeled metabolites.
15:45:40 It's, it's, you know, with our and others have been doing this, I'll show a slide in a second.
15:45:47 But you can basically MIT put a tracer in metabolism and look at the flux how it's propagated through the network, an estimate Interstellar flexes The other type of measurements one can find compared to predictions are the growth phenotypes on different
15:46:02 environments uptake and secretion flexes that are easier to measure and the effect of Jonathan perturbation such as Jeanne deletions.
15:46:12 So I'm going to show just one example of this, which I think is instructive This is from Russ hours lab.
15:46:27 And one of the things I'm going to show this is that what they measure when they do the carbon tracing is effectively the split ratio at specific metabolic junctions.
15:46:30 Oftentimes, one can try to use this split ratio to infer flexes based on also uptick in secretion rates. But what is kind of tricky is that in order to do that you used to use need to use the same stuff geometry that using flex bonus.
15:46:46 So, which is why I think the real the best way of trying to compare experimental measurements with carbon 13 labeling to fluxes is to use the split ratio directly.
15:46:57 And in this paper which showed that actually try to use different objective functions and for which ones. If the best correlation between split ratio that are predicted, and split version that are measured experimentally which highlight another aspect
15:47:10 of this and I know this is, you know, something that
15:47:14 has been brought up and he's often discussed this is a controversial thing. I don't think you know we we may make this assumption about optimal.
15:47:22 But the way I view this is that this is just hypothesis testing you can try different objective functions and see which one seems to best explain if any water organisms.
15:47:32 And there are other aspects of this which is kind of, there are many many extensions there's a whole field now I want to mention too that I think are particularly relevant for communities, and for thinking about the ecosystem level metabolism.
15:47:47 And one this is actually, you know, talking about restricting the metabolic.
15:47:53 The physical space in FBA there is nothing that prevents a network like this to have non zero flux, right so you can have an arbitrarily large flux through a cycle like this.
15:48:05 And this is obviously thermodynamic clean feasible.
15:48:08 One way to get rid of these cycles. So this was actually studied early on by the beer than others, pointing out that you could you what you need to have is a constraint on the free energy.
15:48:21 So for a flux to go for we have a negative free energy along that reaction. So you can express this as this nonlinear constraints between the free energy, and the reaction right in the free energy will depend on the metallic concentration so that's also
15:48:36 a way that metallic concentration can come back into flux balance models in later on. There's been a lot of other really interesting work from axiomatic artist lab, using a mixed integer leaner programming approach to take care of this nonlinear constraints
15:48:52 and Avi and Ron and others have used similar thermodynamic models for studying metabolism in specific pathways. The other aspect that came up already and we discussed is the idea that you know in flux balance and as you don't look at the metabolic constraints,
15:49:10 but not the protein cost in there, there have been ways of taking into account the protein costs, which essentially boil down to additional constraints, where you say that you cannot have an arbitrarily large flux through any sort of reaction networks,
15:49:30 but you have to. There is a total net amount of possible protein that the cell can express. So whatever you do, is subject to this limit on the upper bound and the amount of protein that can be produced in the cell and early on, bag and others have defined
15:49:47 this as a crowding constraints and of course, Terry and others have studied the proton allocation and use this in this cash bar approach constraints allocation FBA.
15:50:00 Yes.
15:50:15 So question about the the thermodynamic FBA, how do you account for the fact that some reactions do are spontaneous but other reactions aren't and are driven kind of indirectly by the other ones do you like model the ATP, or.
15:50:32 And that's a good question spontaneous reaction.
15:50:36 I assume they should be labeled as irreversible in the, what is observed so sometimes right direction is already what was the second part of your question.
15:50:47 Sorry, what was the second part of your question. The question is like some react like presumably building some of the building blocks like has actually like a positive delta G.
15:50:58 It's driven because I, oh yeah on on that, yes yes yes yes so that is taking care of all the couplings are there so if there is an ATP driven reaction that that can be part of the drama cafe because you, you will take into account the overall thermodynamic
15:51:12 balance of the whole reaction, including the driving and the driven. Okay.
15:51:14 Yes, that is, Thank you just say one of the tricky part of doing this around the FBI you need to know the delta G zero.
15:51:20 You know the standard free energy for all these reactions so which makes it hard.
15:51:24 But I think it's a really really interesting Avenue.
15:51:29 So, okay, so let's start moving towards community so how do we extend this models for communities. I want to show you two major parties that have been proposed.
15:51:45 methanogens in this is, to my knowledge, the first paper, using FBA for studying communities. And this was applied to this to organism that are actually seeing traffic exactly in the sense that auto was saying yesterday, one will deplete the hydrogen,
15:52:07 This was done in David styles lab, and I will not go into too much detail I just want to describe how they go from a single organism model to an ecosystem level model.
15:52:10 enabling the other to continue to metabolize.
15:52:19 And the idea is to extend something that already exists in many flex business models for example a nice where you have compartments. So in order to describe convert compartments for example in the East model.
15:52:29 There are cytoplasm metabolites and mitochondrial metabolism metabolites, and when you build a symmetric matrix you have different symmetrical efficient for different metabolites that are just labeled differently.
15:52:44 So it's really just a label simple labeling solution that will allow you to trace in in three different metabolites in different compartments, as if they were separate metabolites then you can add transport across the membrane, for example, I guess this
15:52:58 would be a reaction three will take this molecule a from the cytoplasm to the same molecule in the mitochondria and compartment so this is something that is already built in in a lot of genome scale model so the idea was to extend this idea of compartments
15:53:14 to study communities, and you can take two organisms represented by their own secure Metairie, they may have some metabolites in comments and different metabolites you can just label them differently and construct a big kind of block structures the key
15:53:29 metric matrix that will contain this the geometry of these two different organisms.
15:53:35 So, in this case you would have, you know you could have organised. Having seen traffic interactions, through the exchange of metabolites. Notice here for example this metabolite p that is produced by this organism is necessary for the second organs to
15:53:48 produce biomass so you can easily study potentials in traffic interactions in cross feeding between different organised by building this block structure of the two organisms.
15:54:01 Now this is interesting but it. It has some carrots and I think it was used nicely in in this paper I mentioned.
15:54:10 And I think it has some attention I mean it's fast again it's like FBA you can use it very easily can easily make this block structure, I find it also very elegant and I think it's actually interesting to think, Whether you know the fact that these are
15:54:22 matrices you build block structures. Maybe there is some kind of algebra symmetric matrices one could study, and it is potentially useful but I think it also has a lot of limitations, one of which, which I think is the most important is that you know
15:54:36 if you want to use optimization, you need to assume some kind of ecosystem objective.
15:54:43 And this, you know, could be a big assumption, what do we know people have proposed past this dissipation of energy I think there's no it's very interesting to discuss this possibilities but in practice.
15:54:54 If you do this, FBA communities have to be very careful, and there is something that, you know, people sometimes do but I think one has to be very careful is that you can maximize, you know it might be, it might seem intuitive to maximize the some of
15:55:10 the growth rate of the two organisms and think of this as a good objective for the community, but this will simply just give you know make one of the two probably reach the maximum numerical value with no necessary necessary connection to what the community
15:55:24 is really trying to do. The alternative is to make an ecosystem level biomass, and this would correspond to a new reaction that you arbitrarily put in that says, You know, I know that the biomass composition of the communities 50% Golden say 70% this
15:55:40 30% that, and then you can maximize that combination of the two species, but that's somehow exactly what you would ideally like to predict if you know, use this models to predict the composition of the community.
15:55:49 This is no good in exactly the same way as FDA doesn't tell you anything about the concentration of interest or metabolites. It also tells you nothing about the amount of the species.
15:55:59 Yes.
15:56:12 This is much more serious crime in my opinion.
15:56:18 Our considerations. What you know species are selfish agents they're trying to maximize their own growth. Yeah, and I personally ignore papers for trying to
15:56:29 do that you have a better alternative yeah I'm going to mention that soon but yeah completely, implement, or more.
15:56:42 Yeah, I guess I'm curious why you would assume that it's an additive relationship and not multiplicative if you were doing this kind of thing.
15:56:56 Simply because that's the only thing you can deal with is linear things in FBA, but it's actually it's an interesting which are you could do nonlinear optimization.
15:56:59 I don't know you know it's I mean, I don't know why, a product between two biomass is but, you know, I think it's, it's interesting to think those wouldn't go to zero.
15:57:09 Yeah.
15:57:14 So for this particular system or four systems where species compositions are stable. Right. So then, because we have to have stable, because the composition stable, the growth rate must be equal.
15:57:25 So then, it seems like if you optimize for one of the species is the same, because they will grow.
15:57:34 On average, the same rate. Someone shouldn't seem to optimize the either one or community because they are equal, I mean what do you think of that, uh,
15:57:45 that's a good question. I don't know I'd have to think about this, if they're coupled right things somehow.
15:57:52 Yeah, so I guess. Okay, let me say different I think you're, you're, you have a good point and sense that if there were no cameras that you know in advance the growth rate, or you cannot assume that it's the same growth is that that's what you're saying,
15:58:01 if the species composition seems the only way the species composition can remain the same, is there, the species of going at the same rate.
15:58:08 So in that way that coupled the growth Yeah, which is identical to community growth rate Yes, yes optimize the idea or the community is irrelevant and because of the same thing.
15:58:17 Yes, yes. So I yeah I think that's true I don't know how it affects, you know, thinking about this models but I can remember.
15:58:25 Just briefly censored is giving you a hard time I'll give him a hard time that species may not choose, you know by evolution to grow as fast as possible right they're basically growing to avoid anything else that can invade them or any kind of mutant
15:58:40 that can invade them. And that's not always the same as going fast you may poison your the environment for everyone, including yourself, but make it intolerable more intolerable for the others than yourself.
15:58:50 So,
15:58:53 yeah absolutely all the time. So, you know, there's a lot of spite going on out there as well.
15:59:02 I just wanted to, maybe, if we take the example that auto gave us yesterday with the hydrogen Consortium. It seems to me like maximizing the sum of the growth rates is perfectly fine in that context in this in here is ignoring the concentrations right
15:59:14 like it that that would work in that context, so maybe that's not what they did what they did was they they tried many different proportions of the two species and compare those with the, with the data.
15:59:27 I think that just maximizing the sum is problematic but I'd be happy to hear, you know, more thoughts on why it may be.
15:59:37 Yeah, we should be a.
15:59:44 um, maybe this, this is building off of what net and said, Sega said, Maybe this is another stupid idea is, these objective functions don't work but is there a way like if these species are really selfish Is there a way to think about this ecosystem,
16:00:10 in some sort of Nash equilibrium where you know, either like if you're poised in a way where, you know, one of the species tries to do something funny then. Yes, absolutely. So, yes, so there is work done by posting in my life college camaraderie that
16:00:16 tried the computer Nash equilibrium or multi species community based on FBA and the leakage and so on. I, you know, be happy to tell you more but I think I want to switch to show you the next, you know method because I think I wanted to have the complete
16:00:30 picture then you know I think that there are some good aspects of this despite you know.
16:00:35 But, let.
16:00:36 Yeah.
16:00:37 But I think yes you know for things like game theory approaches, I think this actually might be a you know a reasonable approach.
16:00:43 But let me just point out that there is this completely different approach that opens up a lot of possibilities and I think it's very promising
16:00:53 whatsoever in terms of time it's for right now we're going to keep going for a little bit.
16:00:57 Okay.
16:01:00 Sorry.
16:01:01 Yeah.
16:01:08 So there's this other approach called dynamic correspondence analysis which was proposed 20 years ago. Yeah, almost 2000 and
16:01:14 it was kind of dormant there i think you know people are aware of this, sometimes, you know, occasionally use it but it was proposed as a way of explaining the oxy shifty Nikolai Mahadevan Jeremy doors and Francisco develop this.
16:01:29 And there's actually different flavors of this dynamic towards policy analysis one is a global optimization of the dynamic curve of growth, another is a stepwise optimization which are going to describe in detail, soon.
16:01:40 But it wasn't until fairly later that people realize that this could be useful for modeling communities.
16:01:45 So this I think was one of the first processes of papers that use this dynamic FBA the two models. So let me show you what the idea is
16:01:57 here a second another cartoon oversell but with the idea is that you can take an initial condition and you have an initial amount of biomass for the screen organism.
16:02:08 You have an initial concentration of A key metabolites glucose. And what you can do.
16:02:15 You can compute. More on this soon but you can compute the uptake rates at that given time based on the concentrations in Seoul flex balanced analysis regular flex bonus and what they should be able to give you is the rate of utilization of glucose among
16:02:30 all the different flexes read the rate of growth, and you can build this piece wise linear approximation of the growth curve and the utilization of resources where you say, Okay, I decide have a reasonable time interval and deplete, I'll be within a second
16:02:46 in the, you know, remove the amount of glucose that is taken up based on the flux and the initial concentration, and you can iterate this process for different time steps to produce this piece was linear approximation of the growth curves and follow the
16:02:59 abundance of the tablets in the, in the environment. Yes.
16:03:07 Right. Oh, yes, yes I'll get them to me when you you do need, you know, in essence, you need Nicholas meant an equation for the uptake rates of the tablet so this is essentially a hybrid model where inside.
16:03:23 not get the microphone.
16:03:24 It's
16:03:29 Yes yes yes so when we need you need the kinetic parameters for the, for the transport.
16:03:35 And in those, and you can still I mean, you can still make the assumption that the cell inside is a steady state and riches steady state faster.
16:03:43 So it's, you know, inside it's still a steady state externally, it's dynamic and you keep you follow the change of materials externally, assume it's you assume it's taking us, it's putting out a certain amount of as I was raised constant and so forth.
16:03:57 And that's going to be fixed SRA this dynamic process right
16:04:08 now. Matter of caution fits but then so as soon as gonna be okay with whatever parameters you as soon as cannot just put so much as eye out for taking up glucose.
16:04:21 Yeah, yeah, yeah. And so, so that's office. Yes, and I see that how price our response. Yeah, but this can change in this first time interval so you assume regular FBA right but then you say now I you know the glucose outside has changed so the incoming
16:04:36 flux of glucose has changed right so i recalculate the next step based on the new consumption of glucose and the newly recalculated was one other thing that will change his shots.
16:04:47 Welcome coach level goes down, regulate glucose transporter.
16:04:52 We still totally agree we still don't know anything about regulation, there is no regulation here and, yeah,
16:04:58 Sorry.
16:05:00 Yeah,
16:05:03 Right.
16:05:06 So,
16:05:14 You know you tell me a concentration or something in the environment, and I will tell you I will get in any flux you want or you give me an enzyme with a km, I will just put a pre factor of how much of that enzyme I've got, and I can give you any folks.
16:05:22 I think exactly the same.
16:05:26 Oh I see, I see what is it, yes, yes, yes. Okay. Oh yeah, you have to make an, either.
16:05:33 You know, get have a best estimate o V max, which means yes an assumption on the enzyme you know the amount of the transporter and the Cape Cod and.
16:05:54 And yes, I, you know, we don't know that it's it's a, but you can you know based it on measure and just. Can I try to. Yeah, we assume it's constant with Yeah, yes, yes, yes.
16:06:00 Just try to clarify for one moment, I understand from Daniel that the dynamic FDA requires some maximum growth rate and mono half velocity constant, which then is instantaneously integrated over time and essentially you're making this idiomatic approximation
16:06:14 where you simulate the FBA model at the at the lowest Greenpoint and then at the next one and so on. But the dynamics of those curves have to come from some parameter that you put it, it's not coming out of FBA, right, the, the, the AMA.
16:06:29 Let me see.
16:06:31 The whole point of FBA it's a linear, and you can't meaningfully talk about the single flux. You have to talk about ratio fluxes.
16:06:41 So, okay, right, it's just incorporating still geometry so we should be talking about flux of carbon relative to nitrogen, or something right but not just one thing.
16:06:53 Yeah so so if I if I understand correctly that let me just, you know, it's the problem is not different than regular FBA. So in regular FBA when you set the uptake rate.
16:07:07 It's usually even worse right do you just either, you know, either you measure the flux, and that's fine, or you make some assumption, you know, based on the concentration for somebody like me Kelly Smith and here you do the same and yes, it's absolutely
16:07:21 true and this is a you know a big problem that we don't know if the transport changes, we, we are not. Yeah, we cannot take take that into account, unless we have data for that.
16:07:33 So it's as if we.
16:07:36 Yeah, we, the limitation that we're modeling here is the limitation due to the abundance of the, of the external Metallica.
16:07:50 So, so if I understand what you guys are saying your, your model tells us the how the concentration of the external metabolite is changing. And then there's this question of what's the transporter part of the FBA solution here or that's the, that's that's
16:08:12 We don't know yet. Can you imagine a version of FBA that yeah I can imagine.
16:08:16 Yeah, yeah, yeah I mean, right, if you measure gene expression in principle right you know not easy but I think it's in principle doable.
16:08:26 Yes.
16:08:35 So in the context, you'll be applying to was ecological communities across feeding and all that, there will be multiple potential substrates out there.
16:08:46 You wouldn't know which is the rate limiting right so even apply FBA to, let's say, which medium know it's very difficult. You simply don't know what is rainmaking time.
16:08:58 And then in this case, okay if you say glucose fine.
16:09:14 Have some form of gas, but then if you have like a camera tablets out there. The Wi Fi What are you gonna do.
16:09:15 So, yes, I mean there are there are, I guess you could, you know i think that's that's that's a good point, I don't know.
16:09:16 Make it maybe take measurement take some assumption.
16:09:24 You can imagine using again, kind of
16:09:28 parse parsimonious or total capacity constraints of membrane transporters.
16:09:35 I don't know that you know I haven't thought about this deeply and I don't know that anybody has but I think it's a really good point.
16:09:42 And, yeah, I mean, similar, there are other things like the km could, but but those are not changing dynamic, So, I agree. I think it's a it's a really important point.
16:09:57 So, let me in, I don't know, Boris if I address your question is this is as far as this or
16:10:08 the you know the, you know, assuming we and we will we can talk more about this later, but assuming this.
16:10:22 The know the approximation of a fixed enzyme transport and limitations being dominated by the abundance of the third is reasonable. What is nice about this approach is that, you know, the screen organism a produce, you know acetate that was not present
16:10:29 in the medium before. And if there is a second organism that can only grow on acetate this organism can grow, and you can have cross feeding as an emergent property without having, you know, said anything in advance and without, you know, as opposed to
16:10:46 the block structure model. There is here you can make the assumption each organism is maximizing its own growth rate. So each of the each of them is being absolutely selfish and trying to grow as best as possible.
16:10:57 But if one happens to secret the metadata the other can use that, you know, fair game. And you could see this as an emergent cross feeding property.
16:11:08 This is the slide work as I mentioned that. Yeah, you you need to have information about the update grades in a mechanic meant and kind of equation for the transporters in an absolutely we don't have a way right now to incorporate the enzyme abundance.
16:11:25 But it's in principle possible.
16:11:30 Okay, there is a lot of models out there and this is just to show you know if you're interested in their discrete agent model agent based based models.
16:11:42 Other types of models, this is just a nice review that summarizes a lot of this different approaches.
16:11:47 Using constraint based models to study community metabolism from Jason Pepin slab.
16:11:53 I want to just point out that one of the things we're doing is embedding this dynamic FPA into a spatial structure where we also model, the growth pressure generated by cells, dividing and modeling diffusion and fluctuations, to to do spatial temporal
16:12:13 modeling of communities.
16:12:14 So this is what we call comments, which is a free, open access softer.
16:12:22 And you know I view this as a platform for collaboration you know precisely for trying to add the missing pieces or brainstorming Adora Sanchez and we'll have come have been harder in this building different components and using it for different work.
16:12:39 I will just point out that, you know, from a dynamic FBA in a spatial setting, you can in principle look at the field of different metabolites a different time points on much acid is being produced so much glucose, see the growth rate on the edges of
16:12:55 the colonies.
16:12:58 You can look at fine population effects. This is a collaboration with zero correlation and nutrient dependent colony morphology. And there is also the work of others Auvers lab, an evolutionary process we can generate mutations and simulate multiple strains
16:13:17 arising from mutations and being selected comments is written in Java, but there is no Python in a matter of two books that make it easy to work with it.
16:13:28 So if anybody's interested in trying to, you know, giving feedback and so on. I pretty sure about this.
16:13:36 And to conclude, I want to point out some kind of some of the questions and potential topics for discussions in addition to the one that were brought up.
16:13:46 One is that I think the incomplete knowledge is this one of the barriers for making this models, you know, useful for communities.
16:13:58 And, you know, uncertainty happens on many different levels from the cinema notation specification of the environment the biomass formulation.
16:14:06 This problem of gut feeling and so on.
16:14:09 This is just a review that baby Bernstein and others wrote in our lab to try and map, all the sources of uncertainty is a specific interest was to see whether we can incorporate this uncertainty in probably in a probabilistic way propagating the uncertainty
16:14:24 all the way from genome annotation to phenotype predictions, maintaining all the sources of uncertainty, rather than choosing at each point, an individual model.
16:14:36 And I think that's not easy but I think it's an interesting way of thinking about this.
16:14:41 Um. The other thing which align with suggestion that Congress was putting in the slack chat is to discuss different approaches you know the consumer resource models represent organisers as know having different interactions and do you have a lot of applications
16:15:01 and many of you are working in this area, the genome scale models, try to predict interaction from the internal structure but they have, you know, many limitations and it's not clear that there are scalable to large communities.
16:15:14 So I think it's actually interesting to me that the motivation is to discuss whether. No. Are we really stuck in this two discrete paradigm are there, you know, intermediate kind of models where maybe you can make better consumer resource models by taking
16:15:29 into account the fact of metabolism a change as a function of external environments and so on. And the other thing is that, you know, some people have been thinking that perhaps we can build ecosystem levels while models of metabolism were doesn't really
16:15:44 matter for us, who perform what function you can treat the whole metabolic network of our community as a soup of enzymes.
16:15:50 I think there are a number of issues, but I find it fascinating and discuss.
16:15:57 The other thing that I find interesting is that, you know, we can try to go from organizational models to ecosystem level models.
16:16:06 But many of us I think are interested also the scalability to larger, you know if you try to connect this to ocean biochemical cycle models, how to do that, you know, are there ways of reaching the scales, whether from playing fair in some parameters
16:16:21 from the simple models to plug into larger models. I just find it fascinating that you know metabolism really has no discrete levels or you know it has really covers all these different scales maybe there is some disagreements but it covers all the scales
16:16:36 and I think we are really looking at just some specific skill.
16:16:40 And the last thing I again, you know, I find it really fascinating that in the end.
16:16:46 We know, we know and we talked and auto mentioned yesterday and I think this will you know we'll hear more about Redux and and the fact that electrons power living systems and
16:17:00 and you know in the analogy of an electric circuit is not just an analogy I mean it you know we are big, complicated electric circus where I think one of the key things that is happening.
16:17:11 And this is just my be, I don't know, but the I think the key is that the electrons do very small steps with very little dissipation. And I just find it interesting to discuss I think it would be an interesting topic for physics physics and chemistry,
16:17:26 talk and try to figure out whether it's possible to build a model of metabolism that really focused on the electron flow.
16:17:34 Anyway, happy to discuss this. And I'll stop here and thank you for the questions and have expect more,
16:17:49 I guess, in the spirit of potentially doing an experiment to test like limitations of different aspects of these models. I guess like as a toy, suppose you take these minimal genome, things that people have been mentored as organisms which you strip out
16:18:03 of metabolism right and you then feed them all the known metabolites in the environment. And you ask what is their response rate growth whatever else you want to measure how realistically Do you think you can capture the response of these organisms when
16:18:18 they're growing, and that's it that's a, that's an interesting question I mean we.
16:18:25 I've been talking to people that do kind of synthetic cells and. And to some extent right if you, if you have full control on on how much enzyme there is in the system and you can remove or, you know, Alastair regulation, then I think yes you know things
16:18:42 should work much better and then they doing with the complex a real cells. So I think, I think it's very interesting I think you can imagine going both ways trying to, but I find it.
16:18:53 It's very appealing. Think of whether synthetic systems could give an opportunity to test.
16:18:59 Maybe unrelated question, I guess like I'm not sure I'm understanding the methodology correctly but are there known, I guess in the spirit of trying to include some regulation of genetics or cell biology or their own sub modules that we know self score
16:19:15 core regulate such that you can collapse parts of the network down so that you don't need to actually do this.
16:19:21 Yes.
16:19:23 Yes, that's a beautiful point i you know i there, there have been many efforts to try to incorporate gene expression in flux balance models, some more successful than others.
16:19:33 I think it's hard because there is a know there are multiple steps from expression to protein to flux.
16:19:43 I'm not sure, you know people have thought about correlations of flex is I'm not sure if anybody has done precisely what you would you suggest but I think it's a very nice idea, because I guess their specific example I was thinking what was in the context
16:19:53 of like amino acid production but I know a number of amino acids are covered in very fundamental ways in the way they are produced right it's not arbitrary that you didn't make any combination amino acids you want.
16:20:05 Thanks.
16:20:08 I think there's.
16:20:15 So I have a specific and a more grandiose question.
16:20:18 So, do you have a sense of how sensitive the genome scale models are to miss annotation. I mean, probably from metabolic genes, we do better than we do for other things but still there's a lot of stuff that we just get wrong.
16:20:32 Yeah, and I guess at a at a more grandiose level, I'm wondering whether you think that there's any reason to believe that when we start to combine models, we actually.
16:20:44 Maybe you can like, get rid of some of these noise that you kind of get, you know, you get to average out all the things we don't know hey I'm kind of, you know, get a community response better than you would for an individual organism.
16:20:56 Yeah, so, I think a lot about this because it's as if, you know, everybody's excited about using FBA and all sort communities and suddenly you know everybody wants to build models for the organism and so on and well have no idea I don't even have an idea
16:21:12 how to test this and so on. So, I have two answers. One is that i agree i think there may be answered, you know, questions to address question in a, in a statistical way, in a little bit like what people do with consumer resource models maybe we're not
16:21:26 gonna be able to predict what that specific organs will do, but maybe if I have a representative, you know this stacks and stacks and so let me, there may be viable way of using these models, a slightly different way.
16:21:39 And in the other is that I think more and more about phenotypes in in whether you know it says if, despite sequencing everything now, we're still we're still don't understand, you know, many genes and genomes, but at the same time, phenotype thing is
16:22:10 Sometimes I like to think, you know, do you need genomes, you know, can you reconstruct the models just based on phenotypic data, maybe at some point we won't you know so and this may be a little bit extreme and we probably need both.
16:22:20 But, you know, the transits that I don't know and I you know wander and think about this a lot in case would be nice to have a test case where you somehow specify what phenotypic data is useful and what is not because I think they also do a problem with
16:22:35 two sides not having a meeting of the minds. Yeah, yeah. So that's one of the reasons I got excited again, I mean we did this same string chemistry work many years ago, and recently I started working on this again, precise for this reason I wanted to
16:22:49 generate models where I could have full control I know exactly what's there, and I can generate an arbitrary number of phenotypes with the models and see if I can recover the network.
16:23:02 Sergei.
16:23:07 So let's let's actually go back to this first slide because the Amanda three conclusion slides I wanted to share my thoughts about it, maybe just the next one, the one after this right so.
16:23:20 So the left, sort of picture here is consumer resource model the middle one is the, the marriage between consumer resource models and VH is in my mind what you and others accomplish in dynamical FB because it is really a marriage of be but I guess my,
16:23:40 my comment on this is that there is a limit of what we can do computationally, and I think we reached this limit, so there is no magic bullet that we will apply and suddenly communities will start making sense.
16:23:54 And even if you forget about all the complexity of what happens inside that individual organism which is considerable with all the reallocation of resources, everything which MVA doesn't capture this whereas important enough to just say that let's say
16:24:09 that we believe in consumer resource model can be us, you know, generate a set of experiments and feed all the parameters of the consumer resource model and I haven't seen this done so far convincingly, you know, that the for a well characterized experimental
16:24:28 system was more than 32 species. Here are the parameters.
16:24:33 So I guess.
16:24:35 I guess the summary is that the, what can we learn, I'm bringing it more as a question so either a set of experiments even visit potential of scaling them up massively with some robotics and artificial intelligence what only not what that at some point
16:24:55 we actually reached a point where we can predictively model a community with species abundances when leads Who, who, who are compete schools and so on.
16:25:05 That is kind of my, you know my wife says here.
16:25:10 Yeah, I mean, that's interesting. I think one one thing I can tell you is I know many, you know I hear more and more people want to use machine learning to predict things and so I find it still nice to be able to, to have, you know, the network in there
16:25:28 and see if you understand the mechanism.
16:25:30 But at the same time, you know I'm with we you know we still do models a lot but I'm, I'm going more and more into experimental work myself because I think I agree I mean we need more experiments, this, but ok I will say one more thing.
16:25:57 trying to their microphone go, but I think also we need more theory I think we're I think we're kind of, you know, it's comfortable type FBA models and you know I'm totally fine with all the criticisms I'm not you know I'm not an FBA pilot i think it's
16:26:04 it's a good approach but it's, you know, it's, it's an approximation with all its limitations. But I think, you know, we, I think as a community we should think more, you know what other before FBA was here you know sometimes our example this, we had
16:26:11 kinetic models and and nobody was thinking about FBA. So what's going to be the thing that you know 10 years from today. I think I'll remember when we used to do FBA models you know I'm sure there is you know other part of that stuff there that we haven't
16:26:22 explored.
16:26:24 I mean one thing that's a lesson for me of using these models is that the small models you can get a lot out of right like there's the Corey Coleman model that comes with Cobra.
16:26:33 You can learn a ton from and a lot of it is I mean I did, tons of experiments that were based on that model.
16:26:39 And they all work basically right like that's like some stuff that Ron has developed and assume it's unbelievable.
16:26:46 So I'm sort of curious if we could imagine some dialogue between theory and experiment where we say like what is the right size of model in other words like did.
16:26:54 We have this this diminishing returns issue where like as the model gets bigger, the more errors it contained in the less information it gives you.
16:27:02 So I don't know I has that been explored I haven't seen a paper about that, not. I mean, a little bit the one I know off is, there is a nice work on core models were no reconstruction a lot of like core carbon metabolism models for a lot of organisms,
16:27:17 which is, you know, somewhere in between.
16:27:19 Sometimes people still use the Corey calling models a valuable thing but but no I don't think that's, that's the kind of thing that I think we should explore more in.
16:27:32 You know there are there are in Marina college and people have this mock molecular pool models that are also a little bit of a hybrid, but again that's, that's my point with the discrete you know we have this discrete paradigm.
16:27:45 This is our way of deciding how much complexity you want or need in an in a model based on your goals.
16:27:50 I don't know, but I think, I think we haven't started in exploring this.
16:27:59 This might be a dumb question coming 90 minutes into this talk, but what is the like simplest, what are we trying to model here like what's the simplest experimental like when it came to my mind was.
16:28:13 You want to predict if you change the nutrient conditions, this is for single cells, became the nutrient conditions. How will the ratio of fluxes in metabolic pathways change is that the core thing that we're trying to predict.
16:28:25 Yeah, So I guess that's one of the things we tried to work with regular FPA, I would say that in the community models, right would probably many of us would be would like to have regular as a model that based on the environmental condition will tell you
16:28:43 who is there and what proportions. Right. And what functions you know or and are there multiple equivalent states, I mean there are tons of questions starting from that but i think that's I think one early.
16:29:02 Thanks to stolen. Yeah, I guess in the spirit maybe a serious question in this argument we had the In the beginning it sounded like you know with these consumer resource models one knock against them is that we have to put numbers on all those input and
16:29:14 output rates and we don't know what they are so we draw them from some random distribution, and it sounded like in the dynamic FBA thing that that was also eventually true for the input and output flexes here, some numbers, we won't have to put Yeah,
16:29:28 you have to put those.
16:29:30 have to put those. Yes, but they can change well okay as opposed to the substrate cost, yes yes but some some inherent VSNK. Yes.
16:29:38 Yeah, so I guess I'm sort of curious from from your perspective what what do we sort of gain going from left to middle here like is it. Yeah, yeah. No, that's a great question.
16:29:47 I mean, we started using more and more consumer resource models ourselves because you know they have a lot of advantages they are beautiful convenience on where I get really frustrated right with that is that, you know, simple things like you have aerobic
16:30:00 metabolism anaerobic metabolism.
16:30:03 I don't know, I mean if anybody knows how to capture that in a consumer resource models I'm happy to hear right now i i don't know in there other simple things that are very obvious metabolic things that FDA will do for you and I cannot imagine doing
16:30:28 research. So I guess a follow up question to that so is it a question of like specifically like I know these chemicals and I want to predict when aerobic versus anaerobic metabolism happens, or is it that at the scale of a community you think certain
16:30:34 sizes of communities can appear only if you have those two different types of metabolism and if you don't have that in your model use won't even get, you know, the statistics right yeah i don't i don't know i mean i think that's that's what open questions
16:30:48 Boris. And I guess I'll go back to my question but ratios. So, I think I understand what's happening in the basic flights balance it's the geometry linear constraints.
16:31:01 And you can define some suitable polygon, whatever converts.
16:31:08 But when you were explaining that you're very clear about how this tells you nothing about enzymes and regulation and how you move achieve some particular point.
16:31:24 The bounder of the hall. Yes, right. so then you start talking about this dynamic flat file and you pose a question of, you know how much of what community, there is right.
16:31:33 So as far as I'm concerned community is an answer.
16:31:37 Right. And now you were talking about how much of this hands on there is and how much of that stuff that you do.
16:31:48 Yes, my question is, some magic got this. Okay.
16:31:56 Okay.
16:31:56 Okay, great question. Let me tell you what i. So, you can decide to do this for at any level you want right you can take any reaction of the NSA this reaction I don't want to model this is just as a flux with constraint in and out.
16:32:11 I want to know the Nicholas Menton equation for that reaction.
16:32:18 And you know you, but you, then you have to know the concentration of the metabolites So in principle, you can do it in any level in within within FBA for any arbitrary number of reactions.
16:32:27 You can instead of predicting flexes but it's convenient to do this at the boundary of the cells where you say, I just need to measure the exercise or concentrations, which is, you know, experimentally much easier to do, and I can constrain with Nicholas
16:32:42 men than the flux is going into the cell.
16:32:47 So it's just a convenient level at which you choose to use kinetics, as opposed to FBA.
16:33:03 And
16:33:03 that's why.
16:33:10 right, right, right.
16:33:11 Right,
16:33:12 Yes, yes. But, yes, but but in.
16:33:18 So, it's fine to try and model the whole spaghetti bowl with, you know, people have been talking about measuring every reaction and every KD and everything and putting in the big competitor.
16:33:45 plausible, is, is the output. So I think the two things are not incompatible i mean it's it's okay to have
16:33:55 a steady state assumption, interestingly, and still model the dynamics extra salary.
16:34:02 You know there may be conditions in which that better that's better or, you know, worse but I think that's not incompatible and.
16:34:08 And again, you know, in fact people do this I mean there is this work. I remember who did this when you know when you write the equation as equals zero.
16:34:22 That's the approximation right but you can write in principle, you know SP is equal the zP.
16:34:30 And if you measure the metabolites a different time points there is nothing wrong with doing the dynamics there too. Right, so, you know, we can discuss more later but, you know.
16:34:41 Yeah, no no no that's great that's a great question. I was saying is that it is the choice of what you model, Connecticut is arbitrary but it's very convenient and I think it's reasonable, and we can discuss, Daniel can follow up on that one.
16:34:58 So, Like you here. Yeah. And my dad.
16:35:04 Yeah.
16:35:15 sportsman who I guess isn't there yet to correct me if I remember some told me some time ago that if you take, e coli and minimum media growing the lab, and then you license it and you run it through a mass spec that half the small molecules, they don't
16:35:29 even know what they are.
16:35:31 And it would your argument be that those are all in such low concentrations they don't matter how much you actually believe the basic, the basic starting model of these are the set of reactions and everything else is important.
16:35:44 Even though.
16:35:46 That's a great question and I agree and I and it reminds me I was a marine biology conference at some point and I was showing our universe of all metabolites that has 10,000 metabolites in someone was showing a estimate of the number of metabolism the
16:36:03 ocean, which is tend to the 11 so there is just you know several orders of magnitude.
16:36:07 Okay.
16:36:10 So one answer to you, Daniel is that, You know the alternatives just not to Lucas the geometry of what we know because what we don't know and I don't think that's that's a useful.
16:36:22 I think with the call I have someone gone through and done a quantitative estimate of how much is left out and how much is that like to do with x things.
16:36:31 Sorry I couldn't hear you had someone for Nikolai gone through and done a sort of semi quantitative estimate of how much, how much metabolites your left out and you know what did they contain as far as.
16:36:42 Yeah, I don't I mean I know, you know, sour measured the movement you know excellent acaba la mix of whatever is secured by all under different conditions there is tons of things that we don't know to, you know, to me, the interesting questions about
16:36:56 is you know I yes I'm sure there are genes dysfunction, we don't know and things we don't really call it but we know a lot right so i don't think we'll get really completely new things.
16:37:09 I think there is no probably a lot of promiscuity to me that's an interesting question right how many things can be produced from enzymes that do multiple functions that we don't know about.
16:37:18 So I think these are all really important questions.
16:37:23 So I guess I'm not going to try and be even more provocative.
16:37:27 So I think there's a sort of fundamental problem here that there's not enough work on model organisms that auto was completely wrong. Yesterday, that if you look at the history of knowledge being advanced in biology.
16:37:41 What happens is people define model systems, they compete like crazy.
16:37:47 By competing they check up on each other and that the problem with what we're talking about today is that everyone is so dispersed, both in terms of approaches.
16:37:58 In terms of the physics and math.
16:38:01 And in terms of the biological systems that they study that the chance that we reach convergence without some people picking like a two or three species community where they can experimentally control all the damn things that go on and and test the rigorous
16:38:20 crucible of, you know, theory and experiment, who's right and who's wrong, there'll be another meeting like this in five years, and it will look just like this one.
16:38:31 So I you know one thing that I've heard multiple instances and I don't, I mean I know it's happening there are different groups doing this is, you know, building the microbial ecology equivalent of the call I write a community that you know you can buy,
16:38:57 Right.
16:39:00 So somebody you know I completely agree we need all the communities that are, you know, completely reproducible and that can be used by different groups, I know that you know the the rise of sphere group groups will want to have arises for communities
16:39:17 and marine community will have, but you know that's that's still fine there is nothing
16:39:26 in the gut, maybe a gut