High Level Requirement Specification

R libraries

Reference	Techreq.R
Referenced Use-cases	QtlMicroarray LandscapeGenomics, MicroArray
Dependencies	ExpressionProfiler BioConductor
Champion	StianSoiland
Status	DEFERRED

	Taverna 1.3	Taverna 2.0
Priority	2
Rough estimate	MONTH(s)	-

Overview

Provide statistical analysis tools from R to Taverna, making it easier for biologists who don't know R to do statistical analysis. Browsable list of functions (ie. services), most importantly exposed from BioConductor.

Overall Goals

Short Term goals:

• Implementation of workflows using ExpressionProfiler components where possible, and interfacing with the R environment where no EP component is available, so some R functionality can be exposed as services in Taverna for workflow composition.
• Expose some existing EP components in Taverna so they can be composed using the Taverna editor and submitted to the Taverna engine for execution.
• Documentation on how to expose EP/R components in Taverna, e.g. expressing the components in Scufl and submitting them for execution using the workflow engine
• Documentation on how to use the Taverna Java API and the consumer tool

Medium to Long-term goals:

• It should come with a set of pre-packaged services specific to microarray so it�s easy for microarray data analysis users to get on with their work. For instance, there could be services for performing various QA tests, normalisation, t-test, RMA, clustering (hierarchical/k-means), PCA, annotation.
• Pre-configured library (or set of services) to expose R and BioConductor functionality need to be provided. This is essential because research community shares most algorithms as R scripts/packages for BioConductor.
• It should be easy to add new packages, expose existing algorithms written as perl/R scripts as services. Good documentation and demonstrations on how to do this would be important.
• The performance should be good for microarray data analysis, which involves performing lots of statistical tests on a set of fairly large (~10-200MB) data files. Using XML messages to pass them between services might produce significant overheads.
• By design data communication between services should be minimal i.e. services should be able to pass references to input data for services.
• Monitoring and logging of jobs via a web-based interface and a documented API (Java/WS/etc.) to these services
• A set of demonstration applications should be part of the distribution. This would make it easy for new users to see how to do things in the correct way.
• There should be good documentation (manuals, tutorials, how to etc.) available for users to do their work effectively.
• Service discovery could be improved. It could be difficult for users/developers to design workflows when there is not enough description provided by service providers. Perhaps a minimal expected description could be recommended/required to publish a service in the registry.
• The workflow editor could be improved for usability.
• The timelines or project plan should be published so user community knows what to expect in what form of timeframe and could provide feedback to improve.

Assessment

R support is probably wholely supported by the ExpressionProfiler and BioConductor, which are listed as separate requirements. This requirement is left over the ensure that all R requirements are met by these two.

Affected Components

Taverna workbench, Enactor^?, Feta (potentially Kave and datastore^?)

Key Tasks

1. Examine and choose between the different paths that satisfy this requirement, such as: (3 days)
   • BioConductor and ExpressionProfiler web services could satisfy most needs
   • Locally running Rserve with scavenger that browses registered functions and libraries, allows any function/script
   • Site-local running Rserve, allows only security-approved functions
   • Globally exposed Rserve, but then why not a web service?
2. Introspect BioConductor/ExpressionProfiler/R for automatically generating either web services or local processors (2 weeks).

Appendix

See also BioConductor, ExpressionProfiler, MicroArray and MicroArray.