? Research supported in part by National Science Foundation grants CNS 0521454 and IIS 0612203.
Data intensive applications in Life Sciences extensively use the Hidden Web as a platform for information sharing. Access to Hidden Web resources is limited through the use of prede¯ned web forms and interactive interfaces that users must navigate manually. Hence, the e®ective use of these resources rely on users' rational interpretation of the associated schema and the presented information. Since the computational model for an application is usually in the user's mind, the user is responsible for reconciling schema heterogeneity, mediating missing information, extracting information and piping, transforming format and so on in order to implement the desired query sequences or scienti¯c work°ows. While simple and relatively modest applications can be implemented and executed this way, large scale scienti¯c investigations are often hard to capture, reuse, and maintain without the support of a set of sophisticated tools.
The traditional solution to this problem was to download needed data from di®erent sources to a local machine and then develop customized applications to implement the work°ow in mind by manually reconciling the schema and format heterogeneity. Although this alternative is e±cient, it lacks currency and invites view materialization related complications. A second alternative is to write glue codes, say in Perl, or Java, to connect the remote sites, download data, and run queries . Although the advantage here is increased currency and less maintenance, the disadvantage is the increased cost of needed programming.
In LifeDB, we o®er a third alternative that combines the advantages of the previous two approaches { currency and reconciliation of schema heterogeneity, in one single platform through a declarative query language called BioFlow. In our approach, schema heterogeneity is resolved at run time by treating the hidden web resources as a virtual warehouse, and by supporting a set of primitives for data integration on the °y to extract information and pipe to other resources, and to manipulate data in a way similar to traditional database systems in order to meet application demands. We use a state of the art schema matching system called OntoMatch, a wrapper generation system called FastWrap, and the latest internet computing tools to build LifeDB and to design a query processing engine for BioFlow. We o®er several language constructs to support mixed-mode queries involving XML and relational data, application design using stored work°ows, structured programming using process de¯nition and reuse, and work°ow design using ordered process graphs. We demonstrate the salient features of our system using a substantial set of online examples in real time. Finally, we show that a graphical tool can be used by a novice user to design applications in BioFlow without actually knowing anything about the language. Readers may refer to the lab home page at http://integra.cs.wayne.edu/ for further information on BioFlow and LifeDB.