UniProtToOWL

UniProtToOWL

Model

Given protein XYZ, we generate

• class XYZ (instances are one protein)
• class XYZ_all (instance is the set of all XYZ proteins in the universe)
• class XYZ_isoform for all isoforms of XYZ
• class XYZ_generic (the 'generic' form of the protein, orthology group)

We also generate the following axioms:

• XYZ SubClassOf: XYZ_generic
• XYZ_isoform SubClassOf: XYZ
• XYZ_isoform SubClassOf: isoform-of some XYZ
• XYZ SubClassOf: member-of some XYZ_all
• XYZ_all SubClassOf: { xyz } [XYZ_all is a singleton class]
• XYZ_all SubClassOf: has-member only XYZ [XYZ_all is homogenic; not in EL!]

Why?

Then, we can distinguish the following:

• Every isoform of XYZ has function GO:123: isoformOf some XYZ SubClassOf: has-function some GO:123
• Some isoform of XYZ has function GO:123: XYZ SubClassOf: has-isoform some (has-function some GO:123)
• All generic XYZ proteins have function GO:123: XYZ_generic SubClassOf: has-function some GO:123

How to talk about possibilities:

• Current/standard approach: XYZ SubClassOf: has-function some (realized-by only (located-in some GO:321)), or, for processes: XYZ SubClassOf: has-function some (realized-by only GO:321)
• The problem: not in OWL-EL, will never, ever, scale to size of UniProt!
• Suggested solution: Transform statement in: at least one member of XYZ_all is located in membrane (or participates in process, etc.): XYZ_all SubClassOf: has-member some (located-in some GO:321), or for processes: XYZ_all SubClassOf: has-member some (participates-in some GO:321)
• This is in OWL-EL!

Open questions

• given a DL Query Q, how can we decide whether to rewrite the query (to use the set representation of proteins) or use the class for individual proteins?

Example and code

• Small example conversion of two protein entries in UniProt is at http://aber-owl.net/aber-owl/diseasephenotypes/drugs/uniprot-test.owl
• Github repo [Warning: hacky, buggy]

Todo

• complete conversion:
  • build a complete data model, able to capture all of UniProt
  • decide on annotation properties vs. object properties
  • test conversion of SwissProt and measure reasoner performance
  • convert all of UniProt and test reasoner performance [maybe in the next 10 years]