Related Documentation:

• GENESIS 3 project TODO list.
• TODOs for the first release of GENESIS 3.

GENESIS: Documentation

1 Publication System: GUI Functionality

The examples in this document are based on publications [3][5][4][?]

This first section gives an overview of what is needed, and the technical comments follow in the next section.

1.1 Loading the different Purkinje cell models into G-3

The following sections list the purkinje cells that we use as targets for the implementation.

1.1.1 Passive Segev Model

Publication [?]

• The Segev morphology is available from the neuromorpho database. There are subtle differences to the soma dimensions with the edsjb1994 model.
• The passive properties of this model are accurately reported in Segev paper. Based on these properties, the different versions of the Segev model can be added to the G-3 model library.

1.1.2 DeSchutter: 3 different guinea pigs dendritic morphologies

Publications: [3][5]

• We only have a recompartmentalized version of one Purkinje cell and it is unknown where we can find the recompartmentalized versions of the other two Purkinje cells. The morphologies are available from the neuromorpho database, but naive recompartmentalization of a morphology changes its computational behavior such that exact reproduction of the publication figures is not directly possible. This needs further investigation.
• The Purkinje cell model reported in [3] and [5] is implemented in the G-3 model library. Both the Segev model and the edsjb1994 model can be loaded. The GUI can then display their differences (see below).

1.1.3 Sergio model

Publication: [?]

• Sergio’s model has a different anamolous rectifier. It should be possible to reproduce his model starting the methods section reported in his publication.
• This model’s morphology is identical to Erik’s model’s morphology.
• I am in touch with Sergio about his model, but am still waiting for his response. He said he will try to recover his model from a backup.

1.1.4 Japan model

Publications: [1][6]

• This is reported as a modification of the original EDS model, reimplemented in the NEURON simulator.
• The morphology is taken from D.P. Shelton (but looks different from Shelton’s), there is no report of different dendritic functional regions, some channels have been modified, others removed, and new added. It is really a new model with some components shared with the edsjb1994 model.
• There is currently no possibility to document the assumptions or hypotheses upon which the model is based, ie. comparing the subregions in the edsjb1994 dendrite with the assumption built into the Japanese model.
• We do not have the model scripts.

1.1.5 Different species

We have morphological data for fish, turtle, rat, guinea pig and mouse purkinje cells. We can ask Rachael if she wants to contribute the zebra finch purkinje cells that she has. These morphologies can be handled as passive models (see Segev model above).

1.1.6 The NEURON Simulator Purkinje Cell Model

Importing Neuron models can be made possible. It assumed to take less time than the backward compatibility module because the Neuron script languages are not as sophisticated as the G-2 SLI.

1.1.7 Allan Purkinje Cell Model

Publication: [2]

Allan’s model is based on entirely different concepts than any of the other Purkinje cell models, but a user does not care much about such differences. Implementation of Allan’s model in G-3 is an interesting exercise we should do to check the flexibility and robustness of both high-level concepts and technical implementation.

1.2 Model Comparison

I have divided differently from what we discussed during the last meeting:

1.2.1 Structural Differences between Models

The G-Tube shows multiple morphology, passive models, channel characteristics and channel distribution tables simultaneously.

Both quantitative and structural differences can be visualized using different colors (green color: same parameters / structure, red color: differences in parameters / structure).

1.2.2 Behavioral Difference between Models

Behavioral differences between models can be shown by running simulations that simulate common experimental protocols:

• Responses to current injection traces are shown for each model that is compared.
• Voltage clamp simulations after application of channel blockers.

A library of SSP schedules is used to configure the simulation required to produce these traces (see also below). The G-Tube allows to browse the SSP library and select a schedule for running a single simulation. More details about this library can be found in section 2.

1.2.3 Functional Difference between Models

• Synaptic stimulation under different physiological conditions.

1.3 Publication, Attribution and Lineage

The G-Tube allows to identify who has contributed what to a model, and to identify the impact of each contribution.

The fundamental building block of a publication is a publication atom. A publication atom is a direct component of a model (such as a channel instance, a synapse instance). Each publication atom is contributed by an author.

Every publication atom starts with an equal impact factor. The use-count of a publication atom determines its overall impact. This attribution model obviously gives high attribution scores to Hodgkin and Huxley and to Wilfrid Rall.

The G-Tube shows a table of all the people who has contributed to a model and what model component they have contributed.

1.4 Reconstruct the figures in the papers from the model in real time

A G-Tube publication has a browsable menu of bullet points / narrative components. It allows to explore the publication in depth. Its structure is outline as:

• The first menu item allows to read introductions and links the publication with other publications.
• The second menu item explains the model in detail.
  • It allows to present functions of the neuron using figures.
  • It allows its lineage to be browsed.
  • It allows to run automated tests on the model including a structural analysis and automated validation simulations.
• The third menu item gives access to all the simulations that have been run, and allows to reproduce automatically the figures that are important to the publication. The G-Tube also allows to manually produce figures that are not part of the publication.
• The fourth menu item is a conclusion that links to the previous menu items and other publications.

2 Technical Comments

2.1 Loading Different Models into the G-Tube

1. The model name selection box of the G-Tube lists the names of all the models that can be loaded.
2. Both the G-Tube and the gshell currently work with one implicit workspace, which limits them to importation of only one model at a time. The model-container uses ’namespaces’ as an abstraction for user workspaces. Gshell commands need to be implemented to access the interface to manage the model-container’s namespaces. The G-Tube connects to the gshell over its standard I/O stream connection to use the new commands.
3. A namespace edit box will popup every time a new model is loaded to allow the user to edit the name of the namespace. The edit box suggests a namespace for use.
4. The name of the namespace that was used to load the model, is visible in the model loader selection box.
5. The namespaces known to the G-Tube are also available from a separate ’Recent’ sub-menu in the file menu.

2.2 Morphology Characterization

The following quantities are visible for both models simultaneously in a small table. This table with morphology characteristics is available from a button in the menu ’Model Construction’ –> ’Explore Model’. (this table should be split in two: one for morphology, one for passive parameters).

1. soma dimensions (different between the two models)
2. electrotonic length of the longest and shortest compartments (edsjb1994 has a very very long compartment).
3. total dendritic length, surface area and volume.
4. Number of branch points, average branch order of dendritic tips.
5. RM, CM, RA, ELEAK, number of spines, number of compartments.
6. List of transmembrane currents.

The quantities mentioned above can be computed by the model container and are made available as yaml text files to the GUI from the standard I/O connection with the gshell.

For example .

2.3 Channel Kinetics and Distribution

The GUI shows a list with all the transmembrane currents / channels found in the model.

1. The channel characteristics table shows the gate parameters and the reversal potential for each channel (table 1 paper [3]).
2. The channel distribution table shows the current densities (table 2 paper [3]).
3. Clicking a button in the table with channel characteristics shows a plot of the steady-state and time constant against the membrane potential. This relationship is calculated by heccer and made available to the GUI via the gshell ’tabulate’ command.
4. The voltage clamp current can be made visible too, using a library of SSP schedules (see below).

For example see .

2.4 SSP Library of Simulation Configurations

• A library of SSP schedules with different stimulation paradigms is available. The GUI instructs the SSP scheduler to load a schedule from the library and then runs the simulation. This produces the output for one figure. Tables and channel kinetic plot reconstruction may need a different mechanism.
• The SSP configuration library can be browsed using the show_library gshell command. Each configuration is shown using its description inside the yaml file.

2.5 Importing Neuron Models

• The G-Tube / gshell load SLI models in the same way as NDF models (and necessary conversions are applied in the background).
• Importing Neuron models can work in the same way, with conversions applied in the background. The graphical part of the G-Tube does not need explicit support for NEURON files.

2.6 Other Models

Implementing Allan’s model in G3 will require a profound investigation for how to:

• Connect its dedicated solver to SSP.
• How to do its model specification.
  1. interfacing with the model-container?
  2. setup the solver from the model-container?
• The technical reports that document the implementation workflow can be found in the explanation for how to extend GENESIS functionality.

2.7 Publication and Attribution

• The model-container will define a new parameter ’PUBLICATION_ATOM’ that points to an external file. The external file contains ’publication atoms’ possibly including bibtex references to papers, tagged free text such as abstract and author comments. The PUBLICATION_ATOM parameter is available for all model types known by the model container (cell, channel, etc).
• The full pathname of the external file is guaranteed to be unique in a distributed software system.
• Based on the PUBLICATION_ATOM parameters of a complex model, and by assigning ’attribution scores’ to each model component, the attribution of one or more contributors to a model can be fully quantified.
• Likewise the ’local’ past and current importance of a model component can be quantified by counting its usage from the lineage tree. In a centralized database with peer-reviewed publications the ’local importance’ provides an overall measure of the significance of a model component. The algorithms to compute the importance of a model component shares principles with google algorithms to attribute scores to google search hits. They are also related to spanning tree and coverage calculations in graph networks.
• The gshell’s show_library command has extensions for file types such as ’ndf’ files, ’g2’ files, ao. This command is already supported by the G-Tube. A new extension type ’cite’ lists all the citations related to one model (channel, morphology, or other) and allows its lineage to be browsed.

2.8 Reconstruct the figures in the papers from the model in real time

• A library of SSP schedules with different stimulation paradigms is available. The GUI instructs the SSP scheduler to load a schedule from the library and then runs the simulation. This produces the output for one figure. Tables and channel kinetic plot reconstruction may need a different mechanism.
• The SSP configuration library can be browsed using the show_library gshell command. Each configuration is shown using its description inside the yaml file.
• Most of the SSP configurations required for this functionality are already available from different locations in the source code (eg. the regression tests). After initial creation of this library, the buttons that implement functions for model comparison implicitly select one or more SSP configuration files and run a simulation.
• Obviously figure reproduction can also be done manually from the G-Tube menus. For this, the G-Tube needs an interface to access and load schedules from the SSP library. This interface is available from a menu that is part of the (still to be described) publication / research workflow.

References