>Using RDF, one obvious graph model is to make each cell a bnode of some >type (e.g., "gene expression measurement"), and link it to one column node >and one row node. The result is not directly a list of lists, but a unique >projection mapping of two ordinate nodes: a web of cells to be exact. > >In effect, each cell "knows" that it belongs to a row and a column >identifier. This structure has the added advantage that any algorithm that >processes cell values, can also evaluate the set of values linked to from >each row and column identifier (e.f., GO, pathways, annotations, tissue >states). Most clustering tools currently use non-standard ways of accessing >such info; using RDF it could be standardized. For obvious reasons I've >called these Hypersheets... >
I agree that representing a table as a collection of cell-objects with row and column properties is the most natural representation. We have been developing an integration workbench to assist in mapping schemata/ontologies. A natural data structure for this task is a similarity matrix, which we represent as a collection of cells that 'know' to which row and column they belong. Peter