"
Annotating scientific data:why it is important and why it is difficult", Rajendra Bose, peter Buneman, Denise Ecklund
Some existing annotation systems:
- Swiss-Prot: annotation database produced by specialist curators
- TrEMBL: automated annotations for proteins ( link )
- IBMdeveloperWorks: an annotation is an XML document that is linked to a target data object
Problems and solutions:
- "pointers" to entries in other databases are sometimes used in the fields to provide mappings between co-ordinate systems
- it's easy to propagate the annotations through the operations of the relational algebra BUT inverting these rules is non-deterministic because the annotation in the output could have come from more than one place in the input
- An extension of SQL can be developed to propagate the annotations. One problem is when one attribute from two different tables, with two different annotations in each table, is part of the WHERE clause but not from SELECT clause. In this case, should both annotations be sent ? The solution can be to allow user to control the flow of annotation by adding some further propagation instructions to the SQL query
- For reasons of database security many annotations are stored externally. They require a coordinate system in order to specify how they are to be attached to the data
- External annotations can be retrieve by given the tuple: (table_name, tuple_id, the attribute_name) -- this can be a stable coordinate system
Open problems:
- Many existing annotation system provide only a limited ability to query over annotation values
- One needs to know where the annotation is attached to the base data and perhaps why it is attached. How can this be captured in the database and expressed in the query ?
- Understanding the movement of annotations in the manually curated databases
