Integration of data and processes needs stability of the knowledge domain. This implies "a-priori" deep knowledge of the domain and well defined information and data, both leading to a standardization of information schemas and formats. Also, essential is a clear definition of the goals of integration processes. On the contrary, integration fears heterogeneous data and systems, uncertain domain knowledge, highly specialized and quickly evolving information, lacking of predefined, clear goals and originality of procedures and processes. In biology, a pre-analysis and reorganization of the data is very difficult, because data and related knowledge change very quickly. Moreover, complexity of information makes it difficult to design data models which can be valid for different domains and over time. Finally, goals and needs of researchers evolve very quickly according to new theories and discoveries, this leads to frequent new procedures and processes.

Current integration methods, that are based on syntactical tools like explicit cross-references, implicit links (e.g., through names of biological entities) and common contents (achieved by using common vocabularies, reference lists and lexicons) are inadequate. Instead, new methods based on semantic links, such as those that can be derived by using metadata descriptions and reference ontologies, seem more adequate. Flexibility of systems, including the ability to support frequent changes of data, software and analysis, is mandatory.

Integration of heterogeneous data is anyway needed to achieve a better and wider view of all available information, to automatically carry out analysis and/or searches involving more databases and software and to perform analysis involving large data sets. Finally, only a tight integration of data and analysis tools can lead to a real data mining. In such a context, the need is felt for a system that is able to improve the information accessibility.

Among current ICT technologies, workflow management systems, in connection with Web Services, seem to be the most promising ones. Reasons for the setting up of Web Services in bioinformatics have already been presented [1, 2]. These include the need for overcoming the scaling problem arising from the use of high-throughput experimental protocols that provide such huge results that their analysis needs a "high-throughput" process in order to be studied in an adequate time scale. This could not be achieved through the traditional approach implying manual access to web sites. Instead, software driven access to Web Services implementations of the required analysis software could achieve it. Also, WS would offer bioinformatics the possibility of implementing a really distributed analysis environment, while protecting intellectual property rights for data, algorithms and source code, that would not be copied and would remain on the owners' information system.

WS have already been implemented by many Institutes and service centres in the biomedical field. Partial lists of Web Services for bioinformatics are available at the myGrid Wiki site [3] and in the Taverna web site [4]. Also, Web Services can be retrieved and accessed through the MOBY Central, an archive based on BioMOBY, an open source software that implements an architecture for the discovery and distribution of biological data through Web Services [5].

Web Services alone are not sufficient for automating "in-silico" bioinformatics processes: the notion of workflow must also be included. Workflows are defined as "computerized facilitations or automations of a business process, in whole or part" (Workflow Management Coalition, WfMC) [6]. Their goal is the implementation of data analysis processes in standardized environments and their main advantages relate to effectiveness, reproducibility, reusability of intermediate results and traceability. Effectiveness is achieved through automation of repetitive procedures: being an automatic procedure, a workflow can free bio-scientists from repetitive interactions with the web, at the same time supporting good practice. Reproducibility is also granted by the implementation of repetitive procedures, although it is limited, in biology, by the frequent update of information sources; anyway, analyses can be replicated over time. Reusability is implemented by storing intermediate results and by allowing their use in subsequent workflows executions. Finally, traceability is achieved by storing intermediate results and allowing their analysis: the workflow is then carried out in a transparent analysis environment where data provenance can be checked and/or controlled. This is especially important when unexpected data are obtained.

Workflow management systems should not be compared to other integration systems, such as the Sequence Retrieval System (SRS, [7–9]) since they carry out tasks that are quite different. While SRS is able to perform limited, predefined operations (i.e., boolean and linking operations) on a local set of databases, a workflow management system is able to carry out any kind of elaborations and analysis on remote databases. Instead, an SRS site could be remotely queried through a properly programmed Web Service and its abilities, such as querying more databases at the same time, could therefore be added to a workflow. With workflow management systems, query processing on multiple sources can be achieved by carrying out parallel searches and later merging results. Alternate processing is also available with workflow management systems. This can be achieved by assigning the same task in a workflow to more services, by also providing them priority levels, and by invoking the services having the highest priority level first. Services with lower priority levels can then be invoked, if and when the previously called ones should fail.

Many workflow management systems have been proposed in the biomedical domain. Some of them are add-ons to other tools, some are autonomous applications. Among open source applications developed by academia, Taverna Workbench [10] from the European Bioinformatics Institute (EBI) is the most well known. Taverna has been developed in the frame of the myGrid project [11]. It is able to build complex analysis workflows, to access both remote and local processors of various kinds, to launch execution of workflows and to display different types of results, including text, web pages and various kinds of images. Workflows execution is carried out by an associated tool, the FreeFluo enactor engine. Processors that can be used through the Taverna Workbench include Web Services, either described through their WSDL definition or accessed through a bioMOBY registry, and retrieval of information from BioMart databases [12–14]. Local processors are also included with Taverna for basic elaborations such as simple list or string processing, definition of constant values, local input/output management. New local elaborations can be further defined and specialised by the user that is allowed to create and add scripts by using BeanShell (Lightweight Scripting for Java) [15].

All workflow management systems developed so far assume that end users know all bioinformatics resources they need, especially those resources that can be reached through a programmatic interface, and are proficient, if not skilled, in programming computers and in the composition of their own workflows. They are therefore not viable to the vast majority of biologists and researchers that are normally only skilled in the use of web interfaces.

A portal enabling the vast majority of unskilled researchers to take profit from these new technologies is still missing. A workflow enactment portal should provide end users with a user-friendly, personalized tool where he/she can register his/her personal preferences and interests, easily identify workflows and keep a record of the results. The ideal portal would also be able to enact workflows that are available through all workflow management systems. A first attempt was carried out within the Oncology over Internet (O₂I) project [16] that led to the design of an architecture for a user interface. We present here biowep, a workflow enactment portal for bioinformatics, that is an actual implementation of an extension of design principles defined within the O₂I project. Biowep manages profiles of users, allows for searching in the repository of workflows, supports selection and execution of predefined workflows and allows for storing interesting results. It presents a user-friendly web interface and it therefore is viable to all end users, by also allowing them to take profit from all advantages of the workflow management systems.

We designed biowep, a workflow enactment portal (web based client application, as defined in the WfMC Reference Model), that allows for the selection and execution of a set of predefined, annotated workflows. The system is available on-line [17].

Biowep has been implemented starting from the architecture designed in the Oncology over Internet – O₂I project, but it is:

Of course, such generalization has implied changes in the initial architecture. Such changes have been taken into account during the implementation of the portal.

Furthermore, end users support has been implemented and the software has been made available to interested researchers under the GNU Less General Public License (LGPL) license.

Selection of a workflow

In biowep, users are authenticated. The system stores information on workflows executed by each user and it is therefore able to list workflows executed by him/her. Last executed workflows are listed first and for each workflow executed in the past, its version number and the execution date are also shown.

The system also supports retrieval of lists of available workflows on the basis of the role of the user in his/her organization (e.g., researcher, clinician, computer scientist) and of the domains of interest (e.g., mutation analysis, gene prediction). In the latter case, workflows are listed by date (last executed first), while, in the former, they can also be listed by number of executions carried out by all users (i.e., by popularity among users of the system). In figure 1, the web page listing all workflows available in the workflow repository is shown.

The list includes the name and a short description of the workflows, together with their current version number and the last execution date. In these pages, two buttons are always available for enacting the workflow (button 'run') or retrieving its details (button 'details').

In figure 2, details of a workflow are shown. These include its overall annotation and the annotation of its main steps. Also available is a link to a diagram of the workflow.

Search of a workflow through its annotation

Search and identification of workflows of interest can also be achieved by means of the annotation of the workflows. In figure 3, the web page allowing this kind of search is shown. Conditions can be defined on the application domain of the workflow, as well as on its type (the kind of elaboration or analysis that it performs) and the type of its input and output fields. Conditions can be set on each column (see figure 3 again) and they are then combined by using a logical AND. When multiple conditions are put on the same column, these are combined by using a logical OR. An example query could be: find all workflows in the molecular biology domain (application domain) including at least one elaboration step that retrieves (retrieval task) DNA sequences (output) on the basis of a Genbank accession number (input). Of course, end users are not obliged to put conditions on every field: these can be left undetermined. A search that does not impose any condition on any field will result in a list of all annotated steps and workflows.

Results are listed in the same page and include the annotation. Also included is a note that specifies whether the retrieved data refers to the overall task performed by a workflow or to the task performed by a single step in a workflow. In the former case, the workflow can be enacted, while, in the latter, a list of all workflows including that step can be requested.

Workflow enactment

In figure 4, the input form for the execution of a workflow is shown. In this page, input fields are described in details and suggestions for possible input values are reported, so that the required data syntax is clearly shown. Required and optional fields are pointed out.

The enactment of workflows created with Taverna is carried out, as already said, by using Freefluo. In this case, the execution is performed on the server and results are stored in the system and made available to the user as soon as they are available. If the execution takes more time than a predefined period (usually 30 seconds, but this time can be changed by modifying a parameter in the configuration file), the workflow is executed in background and the user is invited to retrieve results later in the results section. In this case, results are also returned by email. Instead, workflows created with BioWMS, are executed by issuing a request to the Hermes server that is available at the University of Camerino. In this case, results are only returned by email.

Visualization and management of results

In biowep, workflow executions and related results can be saved, either temporarily or definitively, stored and later retrieved, analysed and used for further analyses. In figure 5, the web page listing all saved results and allowing for their further visualization is shown. Results can currently be displayed on the computer of the user by using a java library that must be downloaded from the portal and installed locally. A version of the java virtual machine must also be available and running on the user's side. The visualization library is derived from Taverna Workbench and it includes some extra java classes.

Support for users and developers

Support for users and developers is available in the associated site [23] from where interested researchers can retrieve all available documentation (user and installation manuals, presentations, papers) and download software, database structure and workflows (see figure 6). Archives of mailing lists are also available at the support web site. Three mailing lists have been created and will soon be announced and started: biowep-announce, biowep-forum and biowep-dev. The first is an announcements list for informing users about availability of new versions of biowep and new workflows. The second is an open discussion list on biowep features, also aimed to answer users' questions. The third list is restricted to developers and it is the depute list for discussions about improvements, new features, bug fixes.

Finally, researchers willing to submit their workflows for inclusion in the biowep repository can upload them through the ad-hoc form. Software download and workflows upload are limited to registered users of the portal. So, a unique registration is requested for accessing the portal and the support web site.

We developed a web system that supports the selection and execution of predefined workflows, thus simplifying access for all researchers. These workflows are designed to access and to retrieve data from various Web Services, that we feel are the most promising among ICT tools in view of the automation of network based data retrieval and analysis in biology. The implementation of Web Services allowing specialized software to interact with an exhaustive set of biomedical databases and analysis software and the creation of effective workflows can in fact significantly improve automation of in-silico analysis. Biowep is available for interested researchers as a reference portal. They are invited to submit their workflows to the workflow repository.

Biowep is currently being further developed in the sphere of the Laboratory of Interdisciplinary Technologies in Bioinformatics – LITBIO. Foreseen extensions of biowep include, apart from the addition of new workflows, integration with more workflow management systems and engines. We specially aim to add support for execution of workflows in a Grid network environment.