owltools
owltools copied to clipboard
owlcollab
Proposal: class utility metric
Classes near the leaves of an ontology provide utility by providing specific classification and disambiguation from other entity types
Classes near the roots of an ontology provide utility in grouping other concepts. However, if the class is not acting as a superclass or other kind of grouping (e.g. partonomic) then it is not providing much utility
A utility metric should take into account both ancestors and descendants. A node should only be penalized if it hangs out near the root AND does not group classes.
A naive metric would be max distance from root plus max distance to farthest leaf
Honest question here, but how useful is a utility metric without reference to some kind of annotation set?
It can be useful for detecting dumb classes like this: http://purl.obolibrary.org/obo/UBERON_0023869
which is dumb regardless of annotation
(will be obsoleted in next release)
annotation IC can be used in place of distance but at some point this blends into http://dx.doi.org/10.1038/nbt0210-128
It would also be useful for Bgee: we are trying to rank tissues with expression of a given gene, notably based on expression levels. But it is difficult for, e.g., in situ hybridization data. Having such a score would help.
