Summary of progress during the second through fourth years


Since the first year of participation in the HBP, our goals have evolved along the direction of developing a prototype of the "Modeler's Workspace", which serves as the user interface to the GENESIS simulator and an associated object-oriented database. This system will be used initially by those who construct neural models in GENESIS, and will subsequently be extended to serve the needs of neuroscientists in general. The Workspace allows one to query both local and remote databases for information about models and model components. This information may be used to learn about, evaluate, and compare different models by linking the models to the experimental data which validates them and to relevant bibliographic information. The Workspace will also be used as a tool for modelers to construct GENESIS simulations from model components in the database, and to enter models and their related data into the database. This shouldmake it possible to create sophisticated simualtions with little or no programming of simulation scripts. Our ultimate goal is to allow this interface to be used to explore neural models and related data at any level, from the subcellular level, through single cell models, to network models and systems level models.

As described in the summary of work performed during the first year, we began with a survey of the needs of potential users of the system, and designed mockups of possible user interfaces that would meet these needs. GENESIS is inherently object-oriented, and its representation of neural models possesses many of the characteristics of an object-oriented knowledge base. In addition, GENESIS and its graphical interface XODUS contain many useful visualization tools which are specialized for neural modeling. Therefore, we were interested in exploring the extent to which GENESIS itself could be used as a form of database and user interface. This approach was used in two prototype systems implemented with GENESIS/XODUS.

As a prototype interface for high-level systems models, we created a systems level-network model within GENESIS which includes major projections to cerebellar folium, crus IIa. The model incorporates physiological data on the structure of cerebellar tactile maps as well as maps for the trigeminal nuclei (Principalis and Interpolaris), thalamic nuclei (VPM and POM) and somatosensory cortex. The interface for this model, allowing queries to give the experimental data and literature references which were used to build the model, was demonstrated in December 1995 at the Human Brain Mapping and Modeling conference, in San Antonio, Texas.

We then constructed an initial prototype of the Modeler's Workspace. Although it was clear to us that the final version would benefit from a more general purpose set of graphical tools than those contained within GENESIS/XODUS, the use of GENESIS for this implementation allowed for the rapid development and modification of prototypes, and an easy interface to the tools available within GENESIS. This prototype was demonstrated at the May 1996 Annual Spring Meeting of Human Brain Project Agencies and Researchers.

Progress during the past year (April 1, 1996 - April 1, 1997)

During the past year, we achieved our goal of re-implementing the Workspace in a suitable high-level programming language and connecting it to an actual database system, rather than the set of files used with the earlier prototypes. As planned, we have implemented the interface so that it may be used over the WWW from remote sites.

We have determined that a database with object-oriented capabilities is required for storage and access of the extremely heterogeneous data contained in the GENESIS Database. After exploring various alternatives, we have chosen and used PostgreSQL for the initial implmentation of the database. This is a new, faster, and more robust version of Berkeley's free hybrid relational/object database, which now uses an extended SQL query language.

An important issue to us was to make the Modeler's Workspace easily accessible for modelers around the world. The software should be platform independent, and it should also be possible to access it through the World Wide Web. For these reasons, Java was chosen for the final implementation of the Workspace. The only serious disadvantage we have found with Java is its slow execution speed, in comparison to non-interpreted languages such as C. However, the responsibilities of the Modeler's Workspace do not include performing intensive calculations, as these will be performed by GENESIS, which already contains all the capabilities needed for performing simulations within the Workspace. Thus, we expect to continue using Java as the implementation language for the Modeler's Workspace.

The prototype currently has two modes of use: mining the database for information about models and their components, and building new models using components found in the database.

In the search mode, the database can be searched on the following model properties: cell type, author, channels included in the model, and the year the model was built. For example, a user can ask the Workspace to retrieve all purkinje cell models constructed by James Bower, or all models in the database that contain sodium channels.

Models that match the query can then be retrieved from the database, and visualized in the Workspace. The visualization of a model is organized around an image of its morphology. Rotation and zooming allows the user to get a feeling for what the cell looks like physically.

A set of tools is provided to let the user view different properties of the model. For example, a short abstract or the modeler's notes about the model are available. One tool lets the user see the passive parameter settings for the model, or a specific comparment. Another provides access to the ion channels included in the model. The user can see what channels are present, what their densities are in a specific compartment, and what the channel distribution curves look like.

At any point in the Search mode, the user can choose to copy a part of the model to a scrapbook. The Build mode allows the user to insert these parts into a new model, which can be changed at will and deposited in the database. For example, channels taken from an existing cell model can be used to create a more detailed model, using one of the passive cell morphologies in our database.

Future Plans

Our next step will be to perform user studies on with the present prototype. These studies are meant to highlight possible problems with our current approach to the Modeler's Workspace design - the user interface as well as its present capabilities - and will lead the way to implementing a more capable version of the system. As our prototype may be run over the WWW, we will use this interface to allow selected users to interact with our ongoing interface and database design and provide feedback to our development group.

A very important part of future Workspace development is integration with GENESIS. We want to explore the possible ways of doing this so that we can include support for constructing and running simulations in the Workspace, and making use of existing GENESIS tools for visualization of simulation results and for the modification of models. This will likely involve the creation of a special version of GENESIS to run as a "server", accepting GENESIS commands which are issued by the Workspace.

Another plan for the future is to expand the current database design. The present prototype database can handle single cell models with some auxiliary information. To this we want to add network and system models, capabilities to store various kinds of additional data such as images and movies, and ways to store instructions to set up a GENESIS simulation. We also plan to allow the Workspace to extract information about how a model has developed over time. This will make it possible to view a "family tree" for a model that has been retrieved from the database. The family (or history) tree shows the ancestors of the model. The ancestors are the models from which components were copied for use in this model.

We also plan to do a more extensive evaluation of which database management system to use for the final implementation of the database, and will likely switch to a robust commercial object-oriented database implementation.