Our reasons for using a WWW (HTTP) server as the primary user interface to the ZfishDB, rather than a more powerful X11 based graphical user interface include:
Browsing allows examination of information retrieved by stored links or related by contiguity (i.e. consecutive pages). We intend to use the Illustra HTML data type to store hypertext documents such as journal articles, abstracts, etc. These documents will be loaded from the databases dynamically into the WWW Server providing extensive document browsing.
The recently developed form-filling mechanism in HTML provides a more general interface for implementing a GUI version of the symbolic querying model for database retrieval. We will use forms to construct queries from users which are then processed into SQL queries to the Illustra database. One challenge of this approach is that although conjunctive logical queries (Fig 5) will be easy to interpret, full logical querying of SQL may not be possible (see Challenges, below).
An innovative aspect of the user interface we propose is visual image querying. For instance, the user might ask to see all information on genes expressed in some area of a currently displayed anatomical image, specifying the area by pointing and clicking (Fig 1). Although selecting coordinates within the image is simple (using the HTML clickmap), mapping those coordinates to biological regions (e.g. diencephalon) is non-trivial. Ideally, the system would apply advanced image analysis to deduce the region of interest. Although the Illustra "image" supports features like edge detection, we propose a simpler approach;the submitter of the image annotates it graphically, demarcating the various regions as polygons in an Illustra spatial data type based on simple non-ambiguous landmarks identified according to a standard anatomical model. As the image is added to the database, a spatial map of the region boundaries is constructed based upon these standard landmarks. This map defines the clicked regions which also have text labels that refer to other information in the database [Chang87, Douglas87b]. We will write a dictionary of terms describing this anatomical model which are semantically compatible [Fox94] with those being developed for the human (Carlis letter, Section I) and mouse ([Ringwald94] letter, Section I) databases. If we coordinate design of the zebrafish data model and dictionary with their mammalian counterparts, extensive querying across species may eventually be possible.
Challenges and Limitations. An important constraint that affects interface design is the inherently limited reactivity of the WWW interface. As mentioned above, HTTP interactions are based on discrete exchanges of information rather than on the continuous, reactive connections available in conventional computing contexts. In particular, the interface cannot continuously and autonomously respond to user action; to get a response, the user must click on a link or submit a form. Thus, the interface is limited in its ability to react gracefully to ongoing data entry, graying out items when they become inappropriate, offering context-sensitive help, etc.
Another limitation arises from the distributed character of the system. Because query results must potentially traverse the Internet twice (database-to-system and system-to-user), some delays will be inevitable. However, we have purposefully designed the system to minimize network traffic; the "lazy" retrieval strategy described earlier avoids downloading complete data records until the user specifically requests to view them. We will also use small low-resolution "thumbnail" versions of images for most transactions, providing the user with the option of explicitly downloading the full-size image. With these methods, the response time can be comparable to that provided by most WWW servers today.
Embodying the full power of SQL querying in an HTTP interface designed for non-technical users presents another challenge. For instance, in the interface described in the Solutions section (Fig 4), only conjunctions of constraints on the desired data can be specified (e.g. brain region=diencephalon AND lab name=UOneuroscience AND etc.). This limitation is pragmatic not fundamental; for advanced users, we can provide an interface that allows direct SQL queries. We will need to find the best balance between power and simplicity.
Finally, mapping spatial input to image queries presents a difficult challenge. Several forms of visual image querying have been developed including a) Query by Pictorial Example (QPE); the user formulates a query using a schema of pictorial data in graphic and tabular form [Roussopoulos88]; b) Query by Visual Example (QVE) where similarity retrieval is supported; the user draws a rough sketch of image contents used as a visual key [Hirata92; DelBimbo94]. c) Visual icon-based systems; the user specifies target image contents by placing icons in appropriate positions of the graphic display [DelBimbo92; DelBimbo93]. We propose to explore these innovative forms of image retrieval as future alternatives to our current implementation using the HTML click-map. This effort will be coordinated with development of the human (Carlis) and mouse (Ringwald) databases which have similar goals .
Continue on to the Cost Effectiveness Statement
