CNO internal model
To represent the large diversity of models, we considered that a given model is the results of the aggregation of different building blocks called model component (figure 1). These components should be general enough to allow the construction of any type of models de novo or based on previous models. This allows us to represent the typical case where a new model is build from an existing model with the addition of a new component or the change of a particular component (e.g. using a different cell model, adding a different type of synapse models or adding a model of a particular ionic conductance).
The figure 1 below shows the classes in CNO that were created to implement this particular model.
Figure 1: Representation of models in CNO
The model class is an upper class that represents all the models that could be encountered. Any model that will be annotated is then an instance of this class. The model component class is also an upper class representing the set of components that can be used to build a model. It can be further subdivided into elementary model component and aggregated model component. The elementary model component class includes the smallest components that can be used in models e.g. voltage, current, indices, kinetic rates, amplitude, … The terms contained in this particular class have been for a large part imported from SBO. The procedure used to import these terms into CNO is described later (see “Importing multiple terms from multiple external ontologies.”). These elementary components can be used as variable or as a parameter in the model. The aggregated model component class represents more complex components such as ionic current model, morphology, network layout which result themselves from the aggregation of elementary components and mathematical operators or functions. It is important to mention that the scope of this ontology is not supposed to cover such mathematical functions and operators. The link between the model and its component is made through the Object Property “has part” defined in the [Relation Ontology] (http://obofoundry.org/ro/).
This simple model seems to be efficient to represent any model structure. However a model has also some other important characteristics such as its description (publication, XML file, ModelDB entry, …) or functional properties such as bursting, oscillation, synchronization emerging from the model structure.
To account for these important notions we added two upper classes as shown in figure 2.
Figure 2: Attaching other information to model
The model description class represents the different possible description for the model: publication, a particular implementation, a XML representation, ... As this class is rather broad, we only considered the PMID and the ModelDB accession number in CNO version 0.5. The model quality class represents the properties emerging from the particular structure of the model. This class is rather complex and broad, we chose to focus on populating it in later versions of CNO. To have an example of the type of concept that should be included, you can look at the list proposed on [ModelDB] (http://senselab.med.yale.edu/ModelDB/FindByConcept.asp).
Finally, we created two additional upper classes: defined model and defined model type (figure 3).
Figure 3: Presentation of the defined classes
The defined model class represents the set of models that we could consider as “standard” models e.g. Leaky Integrate-And-Fire, Izhikevitch model, Markram and Tosdyks model, Fitzhugh Nagumo, Morris-Lecar, Hodgkin and Huxley and many more. These different models have well known characteristics and are often used as a base for building up different models. The defined model type is a classification of the different models. It represents the different general types of models such as artificial neural network, spiking network, point process, … The defined model class is linked to the model component class using the relation “has part”. These relations are refined with restrictions to express that one particular model has some defined components. To give an example, the current definition of a Leaky Integrate-and-Fire describes the model with the following components: a fixed spiking threshold, a stimulation current or a synaptic current, a point morphology, a refractory period and a leak current. We are planning to create these restrictions in next versions of CNO.