We Designed the Interface

It was clear that neuroscientists needed more than merely a database for performing one-shot queries. In order for the GENESIS database to be useful it had to provide a means to seamlessly integrate itself into the larger neuroscience research process. To achieve this, particular attention was paid to design a tool that hid the fact that a database was being accessed. Previous database interfaces have either been too low-level, where the underlying schema is exposed, or too high-level so that only trivial queries can be made. Neuroscientists are typically not interested in the underlying data structure used to store the data. They are more interested in finding the data and applying it directly to their research tasks. Our solution, which is an extension of the visualization workspace in our original proposal, was to design the Neuroscience Information Workspace which applies the desktop metaphor (popularized on many graphics workstations today) to manage and organize data extracted from the database, and to relate it to data typically gathered by neurobiologists.

Object-Oriented Direct Manipulation

To support the workspace, our design provided a local database to manage the data that evolves as a result of querying the main database (meta-data). Interaction with the local database is via an object-oriented direct-manipulation interface. Every item of data in the interface is manipulated as an object and each object has its own interface. This allows us to continuously add new object types to the system without significantly affecting the system's interface as a whole. A query agent is introduced to simplify query composition, and a hierarchical conceptual storage metaphor was included to help users organize information that they accumulate over time.

Query Agent and Data Orienteering Maps

Data is extracted from the main database by two mechanisms: by activating hypertext links in a document; or by consulting a context-sensitive query agent (Figure 1). The query agent maintains information on search paths that have already been traversed (contexts) so that it can avoid retraversing these paths in answering follow-up queries (Appendix B) shows a thumbnail sketch of the kinds of traversals possible with the interface). Each data item extracted is either placed in, what is called, a temporary notebook for further exploration; or if the search is a continuation of a search path, will be added as a node in the current notebook. Notebooks provide a means to view the information sequentially as well as on-demand via a local search map (Figure 2). Simply put, this map visualizes the relationships between the search nodes traversed in a session registered in a notebook. More importantly, when multiple maps are merged, the result provides a means to relate information between other notebooks and hence larger contexts. We believe this data orienteering methodology can provide a powerful means to help neurobiologists visualize relationships between multiple levels of neuroscience data. Each local map could represent a subdomain of interest. A combination of maps could represent the relationships between the subdomains. We can use this information for two purposes: to relate information from seemingly disparate research domains; and to transmit this information back to the main database to augment its own data store of relationships for the benefit of other database users.

Information Organization Metaphors

The interface includes a number of mechanisms, borrowed from everyday familiar objects, for information organization. For example, post-its are used to attach annotations to arbitrary surfaces; notebooks are used to quickly collect data; drawers are used as storage for notebooks. From building our data acquisition prototype (see below), we discovered that these differing levels of data organization were needed as a means to enable information sharing between collaborating scientists.

Support for Multiple Data Granularities; Data Exchange; and Operators to Process Aggregations of Data

In traditional journal publications, data is provided at one granularity. Users are then expected to take the data and use whatever is available to them to extract the fragments they really want. In fact, from our initial user study, we were quite surprised to discover that in some instances, neurobiologists would hand-digitize graphs from journals by using a ruler simply because the numerical form of the data was not available. We were careful in the design of our interface, to ensure that data stored in the database could be retrieved at any granularity. Our object-oriented interface and data model allows the user to literally "pick up" any piece of data and export it for use in a number of popular data analysis programs (Matlab, xplot, gnuplot, etc.) Since our database is heavily model-based, one of the data formats will encapsulate GENESIS simulation models which may be exported to a locally residing GENESIS simulator.

In addition to providing such export mechanisms, our interface allows the user to collect a number of data objects and apply built-in tools to them. For example, one could collect a number of Purkinje cell models found on the database, by a number of authors, and apply a comparison tool to them. This tool would display, in a stereotyped format, a comparison of the two Purkinje cell implementations- channel information, morphology, firing data, etc. These tools are maintained in a toolbox, and new tools can be developed and added with time.

Currently these designs have been sketched as a storyboard. These will shortly be presented to the neurobiologist for review before a functional prototype is built. A thumbnail sketch of the storyboard is included in Appendix B.

We Built a Data Acquisition Prototype

We Identified a Suitable Data Model

Concluding Remarks

Return to Brain Menu