 Yes, sorry. So first I have to apologize for the change of topic. I was supposed to present this nice work by Lully and with the assistant of Melanie Stéphane that was published last week in plus one, but it deals with calcium and calcinerine and camcanes too, and you already heard a bit about that. Melanie is with us today, so if you want to talk about this work, which is really great, that shows plenty of interesting things like both calcinerine and camcanes too are actually increasing together rather than the balance, and it's a ratio that changes that you have because of the allosteric property of calcine, you have stabilization, you have a response that outlasts after a certain frequency much the calcium signal and so on, so please go and talk to Melanie. But instead of that I will effectively talk about automatic generation of models, and the reason is the following. So at the moment we have, and that will be the topic of this afternoon's workshop, I guess, you have a huge amount of information about pathways, networks and so on in the literature, and you have also a lot of information stored in database is all over the world, curated or non-curated, many of them specific to neuroscience, but not only. However, in the immense majority of the cases, the modeling process is that one scientist pick a few papers out of the literature, write down a model, either using a processed representation or directly a question, and does its own modeling and simulation exercise. And the question is what about all this information here and here and all the information in the database that is basically there, but we don't use it, we restart from scratch all the time. And I think that's very valid, I mean there is a lot of interesting modeling and simulation to be done that way, but as we gather more and more data we need to have other methods to generate models, so from structural genomics, interactomes, and as Mary already showed, another version of this went from knowledge on pathways. Also, we now are able to build much bigger models, so you maybe have seen this paper in July in Cell on a model of mycoplasma, this is not neuroinformatics, but that's a game-changing paper where people try to model a whole cell, really a whole cell, all the genes, all the molecule, and after the movie of Thomas, even if that was not the right one, I mean I think we're maybe not far from that. And there are examples as well in the literature where large curated pathways give rise to many different models where people started from an existing pathway, so that's the curated version of the yeast metabolic networks published a few years ago, we have now submitted the human one, and for the last four years, plenty of papers were based on this where people pick part of the network and build models. So, because I will mention models and pathways a lot, I need to define what those are. And I will base myself on Wikipedia, as we saw yesterday, Wikipedia is a good source of information, a reliable one. So Wikipedia tells us that in biochemistry, metabolic pathways are a series of chemical reactions occurring within a cell. And that will be our operational definition, you have plenty of pathways databases, but the important thing is that those pathways are detailed representation of reality based on observation, they are meant to represent the reality. But when you do modeling, this is not always what you want. And Wikipedia tells us that the model is just a description of a system using mathematical concept and languages, so you have variables, functions, constraints to limit this model. And the important thing is that this model is an abstract representation of reality, right? Based on needs, you develop your model based on the question you ask and the data you have to generate the model and validate the model and that was already mentioned this morning. So the whole project, the idea was to start from pathways, possibly in proprietary format, transform them in standard format, either to represent the models for a computer or for a human being, generate mathematical models and put everything in a database for everyone to be able to use it. And I will, so a bit of SBML and already some people in the room maybe say, oh no, he will talk about SBML again. That will be just a few slides. So SBML, the symbology marker language is just an XML language to encode models, right? And then you can exchange them. It's very simple. You basically define your variables, the compartments, the molecular species you put in there, the parameters and all the relationships between those variables. You can represent values, so containers, pools that react within containers, between containers. You can put discrete event, arbitrary rule and so on and so forth. So that's a micro-SBML model with two species, ANB, right, in the compartment cell, one parameter. And then I can describe a mathematical reaction, a kinetic law here, regulating the velocity of the reaction. And if I load that in value software, for instance, copasi or cell designer and I click run, I get the same result. And that was not possible before SBML. So SBML now is the level three. And the important thing is that in level three, you have a modular language with various packages covering different aspects of models or different type of model. And today we will talk about the core package, which is shared between everyone, the layout package and package to encode qualitative models, logical models, patronet and also flux balance constraints. The other standard we will use is SBGM, the system's biological notation, which is a set of three languages to represent biochemical, biological pathways. And I won't deal in the difference between those three languages, but you will see why we need them. And I will present an example based on keg. So we did actually the same workflow on other database. We just published the NCIPID database to model transformation. But keg was important. So the Kyoto encyclopedia of gene and genomes has a subpart called keg pathways. If you organize a course in systems biology, the first thing the student want to do is to go to keg because it's nice graphically and download keg information. And the second thing they do is to convert keg information from this proprietary format that cannot be used anywhere into SBML. So that's why it's an illuminating example. So you have two types of maps in kegs. If you take a metabolic network, for instance, the TCS cycle that is very, very important for the brain, remember, brain, 2% of your mass, 10% of your energy consumption, I think. And so in those maps, you have those bubbles here that represent chemicals. And those chemicals are transformed into other chemicals using reactions. And this annotation is an enzyme classification number. And we have another way to represent pathways in kegs. If you look at the signaling pathway, you have a complete different representation where you have those rectangles here, annotated with something that vaguely look like a protein name. And those have arrows connecting them together. So this is a B-parted graph, right? You have two types of nodes. This is a graph with only one type of node. This is a positive regulation that will be a negative regulation. Okay? So it's important to understand the difference between those two graphical representation. It's a fundamental piece of information to understand systems biology. The first representation is a process description. It's a classical biochemical representation used for the last 70 years or so, where you have two pools of molecular species that are transformed into other pools through processes. And those processes can be regulated by yet other pools, okay? B-parted graph. These nodes never connect to this one. This one never connect to this one. So those are representation directional. They are sequential. You can start from somewhere and somewhere else. They are mechanistic. You know exactly what's happening. But they are subjected to combinatorial explosion, something alluded to by Thomas. They're used a lot in process modeling for biochemistry and so keg metabolic pathways are in process description or reactome for those of you knowing these database. And we will encode that in SBML core. The second type of diagram is an activity flow. So this is directional as well, sequential as well, but it's non-mechanistic. When you have that, you don't know what RAF does to make. RAF stabilizes an active state of MEG, stimulates the expression of MEG, inhibits the degradation of MEG. You don't know, right? So you just know that RAF. So this is actually an activity. What you represent here is RAF activity, stimulate, make activity in some way. It can be used to build logical models. And it's used a lot for signaling pathways, kegs in any pathways, but also STKE, for instance, use the activity flow. And we'll encode that in SBML core, the package used to encode logical models. This is a complete workflow. It's a bit more complex than the previous one. So we'll have three different sub workflows. The first one is to take all keg metabolic networks, convert them into SBML core, represent them graphically with the SBML layout package, and converting them into SBGN for the graphical representation, transform those into chemical kinetics models, put them in a database. A second workflow will take the non-metabolic pathway, the signaling pathways, basically, mostly. Enrich them with information coming from other databases we used by Okarta in this example. Generate SBML core, same, we generate the graphical representation with symbology graphical activity flow. And generate logical models, distribute them. And finally, we'll have a third branch, when we will take the metabolic networks in keg for a given species, reconstruct all genome metabolic networks. Enrich them with information coming from Metasite, which is another database of metabolic reactions. And generate models called flux balance constraints model and distribute them. I will start with the whole genome reconstruction. So we take the information from keg, we extract it, we convert it into SBML, and we annotate them with so-called identifiers.org URI, basically URIs that identify exactly all the elements of the model. We do the same for Metasite. And then we will go through the workflow with those SBML files and identifiers that will be all handled, that will be able to relate information. So we merge the information from keg, we protonate on both sides, we mass balance the equation, and then we merge them. The information coming from keg and Metasite, thanks to the identifiers. We add the transporters, you will see in a minute why it's important. We add the biomass, and then we generate the FDM models. So what are flux balance constraint models? The idea is that for any metabolic networks, you can represent a matrix which is a stoichiometric matrix. You have on one dimension all the metabolites, on the other dimension all the reactions. So this is a small metabolic network, we have a small matrix, minus one means it's consumed, one is produced, and zero means this metabolite is not involved in this reaction. So the first thing for FBA is we need to add the boundary condition, the input and output, so that inflates a bit our matrix. Once we have that, we'll go further. So for each reaction, we can have a velocity, and so we have a vector of velocities that we will work with. We don't need the kinetic rate load for FBA. The only thing you need to know is there is a velocity number. And then you do a steady state analysis. So the steady state analysis is basically the multiplication of the stoichiometric matrix by the velocity vector equals zero. That gives you all the possible steady state of your network. But that's not enough because generally our model is under determined. So you can't determine all the velocities just with your information. So you need to add objective function. For instance, you will say I want to maximize the production of ATP, or if you are brewing beer and you're using saccharomyces, you say I want to optimize the production of ethanol. And then you can solve that. So that will be the result of the steady state analysis for those three velocities. If you put your objective function, you can reach a given vector. So now for individual pathway, I said we will generate chemical kinetics model. So we do that with a workflow based on something called keg translators. So we generate SBML core. We clean the reaction using the keg API that allows us to interrogate all the different databases of the keg constellations. And then we will do two things. We will interrogate Sebiore keg, that is a database of chemical reactions. If we have mechanistic rate law for this reaction, we reuse it. This is unfortunately not the case for the majority. So for the majority of the case, we will use modular rate law. So modular rate law is something that's been presented by Wolfram Liebermeister in the group of Edaclip in 2010. And this is great. This is the absolute general rate law that represents any possible rate law for a chemical reaction. And then you have variants. For each of those bits, you have variants. And so for instance, here this will be a generalized representation of mass action law. And based on the data you have, you have to decide the granularity and the generality of your rate law. And for the signalling pathways, remember, we don't have metabolic networks. We just have activity influencing all other activities. We'll use logical models. So logical models, briefly, you have different activities here, A, B, C. And you put logical rules, right? If A and B then C, if C not B and so on and so forth. So you end up with vectors of activity. Here I use a Boolean, but some people use multi-valued models as well. So if you have 1, 1, 1, then B is inhibited by C. C is 1, then B becomes 0 and so on. And you can generate a state diagram and identify the values, behavior of your network. Right. So we release the first version of the models in May with the 22nd release of bi-models database. We have so big numbers. We have plenty of models. Now we have about 150,000 models that represent, I think, a bit more than 10 million reactions. And importantly, they are annotated with 400 million cross references that allows you to identify specifically which metabolites, which and so on and so forth. And if you go to bi-models database, there is now a new branch called pass-to-models. And you can explore them, browse them or search, explore them with taxonomy as well. That's a non-tri in the database for the Glutamatergic Sign-Up. You even have the representation coming from KEG. And I would like to end up here with acknowledgement. Those three chaps have generated most of the data. So Neil was in charge of the metabolic reconstruction, the vulginum metabolic reconstruction, clements of the kinetic chemical kinetics models, finia of the logical models. Tobias took care of the graphical representation of the process-based models, Michael of the logical models. Andreas was the PI on the Tubingen website. Camille is the coordinator of bi-models database and all associated activities. But actually we had many, many more people involved in this work because it was a pretty large work. I won't enumerate all of them, but they will be in my presentation. And I would like to end with an advertisement for a book. So for the few of you who are not authors of this book, all the others are probably in the room. So systems biology applied to neural analysis system is a very, very wide topic. I believe in this workshop we will address mostly biochemical systems. But we have many, many more systems at different levels. So there you will find information from transcriptomic and proteomics all the way to very large network and modeling of EEG and so on and so forth via classical spacking neurons or multi-compartment neurons, etc. Thank you.