Some notes that I have about this:

PSICQUIC is planned as a common web service API for all participating databases. Matthias Oesterheld demonstrated an initial schema for this tool, based on the HOBIT experience of document style web services. As previously discussed, this will be a leveled service with Level 1 supporting simple queries such as get all interactions by organism (TaxID), by interaction property, by experiment property or by protein list. All attending databases would be capable of supporting a cpath implementation of this model, others may need a wrapper web service. It was decided to change protein list to interactor properties and extend the search to encompass short name, aliases and cross-reference. Searching by TaxID would encompass all children of the term used. Negative interactions would not be searched in initial implementations. Level 2 could incorporate more complex queries and feature properties searched by both InterPro cross-reference or MI type.

Returns would be in a document, the maximum length of section returned could be set by the operator, by generating a start index and length. An additional method on the web server could then retrieve the desired number of interactions in batches. A test procedure could pick out specified entry numbers allowing the user to skip large datasets. The possibility of adding a flag to the query to give the option of a URL at which a compressed complete file could be accessed was also discussed.

Complex queries would not be too difficult to establish, however may prove too great a server load. It was decided that both levels would be published simultaneously, with Level 2 being completely backwards compatible. Implementers may then decide which level to install. The first query will send a user query. The return will include a query ID and the number of results. A retrieval function will detail query ID, count, start index and length. Query send may be adjusted to size of client server. Standard error messages will be incorporated. An initial implantation should be available late 2005.

Comments

First Draft by Lukasz

Michael: STRING and also IntAct? may have many, many interactions for any given query, so a "give me all interactions" query isn't feasible. STRING has the notion of confidence levels that could be used as a cut-off, but I think this is too specific for a common query interface. So perhaps it would make sense to add two parameters: limit and offset, with defaults set to something reasonable (e.g. 25 and 0).