The document from the working group is here:
From my perspective the key requirements are:
(1) File annotations must continue to be linked in their life cycle to the files themselves. (That is, the workaround of putting annotations in some other data structure runs the risk of the secondary structure becoming stale as files are created and deleted.)
(2) Data curators need to run queries on the multiple values given to a single file and annotation name. Examples:
(2a) Let's say each file has an annotation 'disease' and an annotation 'subjectId', where each file may have multiple subjects. One may wish to ask, "select entityId where disease='ra' and subjectId='101'" to find the files with data for subject 101 who has 'ra'. The query can't be run against a triple store representation of the annotations.
(2b) count query: "select count(subjectId) where disease='ra'". Again a triple store representation of file/annot-name/annot-value wouldn't allow a simple table query to answer this question.