 In this short time that I have, I'm going to talk about one very specific project within that community. Many of you might know that I work on Neuromal, I work on a lot of other things, so hopefully you might come by my poster. I'm also going to do a poster, and we can talk about some of the other things that I do. But today I want to talk about NeuromalDB, which is a database of Neuromal models. Of course, I'm going to tell you what that is and why this database is helpful to our community. And hopefully tell you a little bit about how it interacts with other resources in our community. And I want to check the time, make sure I don't go over. All right, and also I want to acknowledge some of the people who work on this database. And so the primary person is Justice Brighillis, he's a PhD student in my lab. Also, if I have time, I'm going to talk a little bit about some data analysis that we're doing related to these models. And some of that Virgil and Russell are also students in my lab, and they've contributed to that. And then Rick Gurkin, many of you might know him, he's here at this meeting. He's a major collaborator in my lab, we sort of share a lab. And he's worked on this as well as many other projects that we work on together. Okay, so here's this word fair, you guys just heard all about it. And the only thing I wanted to point out here is that you can think of models as being a kind of data, right? So once you create your model, you want to share it, you want people to be able to reuse it. And we really want models to be also shared in a way that's fair. You know, when you think about data, you might not think about the fact that we need this to apply to our models as well. And so hopefully I will convince you that this project and the ecosystem of tools that we work on, our main goal is really to make models fair. You know, we want them to be, we want it to be easy to find them, we want it to be easy for people to share them in a way that they're very accessible, that you can reuse them, you can evaluate them, you know, things of that nature. Alright, so I need to tell you guys a little bit more about Neuromal. I'm going to fly through this. Some of you may already know what Neuromal is, but if you don't, at least you'll have a little bit of a sense of what it's about. And so I guess a thing to think about is that models have become very complex and complicated. You can't share every detail about the model in a publication, right? You need a way that's machine readable that you can share all the information. I can't just dump out a bunch of equations. That's not going to be helpful. People need to know what the things, the variables, the parameters, what do they represent, what are the units of the parameters. They need to know semantic information about all the things in there. You know, and so this is what Neuromal is all about. It's a little bit verbose, but it's a way of putting all that information into one file and sharing it in a way that the file format, of course, is XML. It is standardized. We don't necessarily think of Neuromal as a standard because it's easy to have it interact with other similar efforts, right? But in some way it can be a standardized format so that a lot of different software packages can sort of deal with the same format for describing models. All right, so very simple example. Let's suppose we have a relatively simple, widely used model like the adaptive exponential integrating file model. All right, so we have equations for that model. In this case it's very easy to share that model. We can write down the equations. We can share the parameters, but I wanted to do something kind of simple as an example. And a key thing here is when you change the parameters, you can get a lot of different kinds of behavior out of this model, different kinds of neurophysiology when you do simulations. So in Neuromal, this type of model is a component. You can think of it as Neuromal knows what this thing is, what this kind of model is. So in Neuromal, all I need to do is share the exact specific values for those parameters. I can think of that as being a model. I can have all sorts of information associated with it. If I want to hear in this, the thing labeled B are two examples of models, this type of model but different instantiations with different parameters. If we run them, we get different kinds of behavior like you see in C. An important aspect of Neuromal, which is not so important for this talk, but just in case you were confused, you heard about it, you're confused, is that underneath these high level components, which are quite modular and oriented around semantics and objects in neuroscience, there's this layer that says, okay, here are the equations underneath this model, here are all the parameters that you need, here are typical ways of thinking about the units for these parameters so that we can do automatic unit checking and all this kind of thing. So an important thing for the database is that Neuromal is quite modular. So you can think of these models around neuroscience concepts. So you have channels, you have synapses, you put these in the membranes of cells. These cells, you can have descriptions of morphologies if you have a multi-compartment model for your cell. You can also have abstract models like the one I just showed you, which we might think of as being a point in space or a patch of membrane. And so all these Neuromal accommodates all these different visions, these different kinds of models. And then of course you can have networks of these cells. And so with Neuromal, because these things are very modular in the model, let's say you wanted to change a channel, you can swap it out with a different kind of channel very, very easily without having to go through some huge amount of code and figure out where does this channel exist everywhere in this code. You just change that channel by changing that one little module in Neuromal for that channel and you're done, right? Okay, so there's an ecosystem of tools around Neuromal. This is just a subset of them. Things like simulators, databases that use this format, visualization tools and things like that. I want to mention one that is very important, which is open source brain developed by my collaborators lab. So Angus Silver is one of the major collaborators on Neuromal and open source brain comes out of his lab. And so the thing about open source brain is it's a platform online for developing models in a very collaborative way. There are models there that are already published, you can interact with them, but it's also a place where people can come together and work on new models or even an individual can work on new models and there are tools there for helping you do that. It's really based on Neuromal and what Neuromal brings to the project is the ability to easily develop tools because you have this standardized description for very complex models. At open source brain you can run models, even very, very complicated models through a web browser, sort of server side, they run it and then you can do visualization, you can interact with the model there at open source brain. This project NeuromalDB is very complementary to open source brain, it has like a different purpose. So whereas open source brain has these projects that are constantly changing, it's built on GitHub, NeuromalDB is more about a model that you think of as being fixed, it's maybe a published model, the models are associated with publications and so we want to have very efficient quick search, you can go there and do a search, we have a keyword search but it's also based on ontologies around neuroscience, so for instance if I search for something like olfactory bulb, then the ontology knows these are all the cell types that you find in the olfactory bulb and the search will bring up models that are related to that. So if you click on one of these models, then you'll go to a page for the model, an entry page for the model, and it links out to things like the publication in PubMed where you would find more information about the model, the authors, provenance of the model, but it also links out to other resources where you can find the model. So you can go in, if you want to, this could be an entry point, you find the model and then you go and look at it in open source brain, maybe you find that other people are working on new versions, new models building on this model. It also links out to ModelDB, which is another resource for sharing models in neuroscience, but where the code for these models might be a lot of different things, you know, not just NeurML. So the models that we have in the NeurML database, they're all also in ModelDB, but the difference is, you know, we're just a subset of there because we only have the models that are in NeurML. Or you could go straight to GitHub and get the model, you know, from there. But the advantage of the database is that we can quickly give someone the NeurML documents associated with the model. Also for every model we give neuron code for the neuron simulator so you can, no matter what language it was developed in, because NeurML is this common format, NeurML is very supportive of code generation. So you can take a NeurML document and we have libraries where you can create code for running the model on neuron, on genesis, on moose. You can write out MATLAB code, Python code, whatever you need. As long as that simulator supports that model, we have mappings to, I don't know, about 15 different simulators and languages where you can run these models. And then we have some other versions associated with visualization of models. If I go to a page for additional information associated with this model, all these models are in NeurML so it's quite easy for us to run, you know, run batch and share information that fully characterizes the models. So you can quickly look at the model and for a large range of protocols that people would usually use to characterize, for instance, a neuron in an experiment, we run those same protocols on all the cell models in the database and then we have these interactive graphs where you can click through all these different protocols and see what this model will do for that particular type of stimulus in a simulation experiment. So you can quickly evaluate the models by looking at what they do. What do the simulations look like? And then the same kind of thing for morphologies. We have these animated gifs showing a simulation in the morphology so you can have a look at it, see whether it's complex or not. If it's a simplified morphology, you can look at parameters related to the morphology and things like that. You can look at other properties of the model, something about how complex they are and how long they take to run. Okay, so I'm already running out of time. I don't have a lot of time to talk about some of the analysis that we've done on these models but because we can run all these models easily and do whatever we want with them because they're all in NeurML, it makes it very easy to run them. We have done some things like look at all the 1,500 cortical models that we have and do analysis based on those models or compare those models to data. So I'll just use this as kind of an advertisement for my poster and you guys can come by and learn about things like how we have clustered models based on electrical behaviors to try to say something about which models fit well into clusters, which models are outliers. Maybe that model was designed to do something different but we can dig into it a little bit and see, maybe it's a way to evaluate models and figure out, do I want to reuse this model or do I want to choose a different one? Similarly, we've done some work to try to compare cortical models to cortical data from the Allen Institute and so similarly, again I don't have time to go through this but if you come to my poster I'll go through all the details where we're taking data and models and trying to say something about how well the models that are out there that people are using fit the landscape of data that we have for cortical cells. So I hope that I've introduced a little bit about Neuromal and NeuromalDB and maybe you see a little bit about how they fit into an ecosystem of a lot of different tools and resources, many of which I didn't talk about at all today, that help make models fair and make it easy for us to share them, to reuse them, to evaluate them and things of that nature. And so Neuromal has had a lot of people work on it over the years. These are people who have interacted in developing the schemas or come to workshops and things like that. And then of course my main collaborators I mentioned, Angus Silver and poor Gleason who works with him at University College London. I also want to thank the Neuromal editors. We have a board of elected editors who help steer Neuromal into the future and work on extensions to Neuromal. I mentioned the database so Justice is my student and Susanna is a collaborator of mine at ASU and then for the model analysis multiple students and Rick also worked on that. So thank you for your time.