Ontology meeting 2016-09-29

From GO Wiki
Revision as of 07:08, 30 September 2016 by MC (talk | contribs) (added concerns about ontology call merge)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Attendees: Paola, Tanya, Harold, Melanie, DavidOS, DavidH, PaulT, Chris

Regrets: Moni, Jim

Minutes: Tanya

Follow-up: Reinstate daily report of terms and stats

Ticket: https://github.com/geneontology/go-site/issues/217

The ticket says that this is fixed, but no one has seen the report. Chris, any news on this one please?

 Some progress. See ticket for details.

Follow-up: Modified proteins

See background discussion here: http://wiki.geneontology.org/index.php/Ontology_meeting_2016-09-15#Modified_proteins

At the time of writing, comments are still missing from some ontology group members. Please review the document here and comment, even if it's just to say that you agree with everything.

https://docs.google.com/document/d/17U1TD3n_QI9coCycJcOUBDa2nRf3ggPR-tKjDMa_ff8/edit

Then Paola will wrap up (or we could do it as a group) and we’ll present, whether it’s at an annotation call, by email to the wider group, in person at the GOC meeting, or all of the above.

 Chris needs to comment and then Paola can wrap up.

EC2GO annotation problems

Current EC2GO annotations are done using the xrefs from the GO ontology file. Main problem occurs if EC number contains one or more dashes, if a gene product is associated with an EC number (from UniProt files, Trembl record for instance), then it becomes associated to ALL of the GO terms that have that match the non-dash numbers.

 http://geneontology.org/external2go/ec2go

What generates these annotations? Need to track down the script. Possibly one written by Chris, Heiko, someone else? This problem should be corrected at the source.

 Archeology:  Found email thread titled '[go-ontology] Ec2GO mappings' March 9, 2015 - Harold raised issue, Chris 
 generated some reports. 
 go-ec-dash-no-generic-ec-parent.txt -- cases such as the one above where we have a dash xref but no corresponding 
 generic xref
 go-ec-dash-inconsistent-with-generic.txt -- cases where the dash xref and generic xref exist but the GO term for the 
 dash does not sit under the GO term for the generic
 Also, http://wiki.geneontology.org/index.php/Ontology_meeting_2016-02-25#ec2GO_Clean_up
 Action items from there:
  AI #1:  Chris will remove ALL EC xrefs in the GO file that have a dash in the number, look like this
  xref: EC:2.-.-.-
  xref: EC:2.1.-.-
  xref: EC:2.2.3.-
  AI #2: Review 250 cases where there are GO terms with more than one FULL EC number (eyeball they are in different
  categories). Harold will look at these, Chris will send.  (HJD and DPH have 1100 enzymes to add for Peifen.)
  AI #3: Do NOT add EC terms without dashes to new terms.

RHEA update request

Where are we?

 Chris went back over his plan presented at the Bar Harbour meeting in 2013.  This attempted to hack representation of stochiometry 
  and bidirectional relations into full logical definitions staying within OWL-EL.  On reviewing these seem overly complex and may be 
  unsafe (they may cause incorrect inference when the inputs/output of one reaction are all present in another).
 We decided to simply add subclassing axioms rather than equivalent class expressions.  There are two ways to do this: 
  1. Keep bidirectionality:  using a pattern that translates the rxn   X + Y -> A + B  into 
      SubClassOf has_participant_collection some ((has_component some X) and (has_component some Y))
      SubClassOf has_participant_collection some ((has_component some A) and (has_component some B))
  2. Encode directionality reflecting how these reactions are conventionally drawn and their predominant direcion under physiological conditions.
     i.g. Translate the rxn   X + Y -> A + B  into 
      SubClassOf has_input some X
      SubClassOf has_input some Y
      SubClassOf has_output some A
      SubClassOf has_output some B
 After discussion, we decided to go with the latter, although review later if we run into inference problems (e.g. from use in LEGO to mean opposite direction). 
 This approach maintains OBO compatibility and as easy to work with. We should probably add a standard comment to all terms explaining the design decision.
  Note: the first approach may also have its problems: reciprocal reactions would have the same axiomatisation.
  Neither approach => us autoclassification, but we may be able to find cases (e.g. transporters) where we could come up with full logical definitions that work.
  Where does classification come from?
  We agreed that we should extract classification from EC wherever possible.  We can do this for Rhea reactions wherever they have a mapping to an EC 
  - or an EC used as a classification.
 Open question: should we also create an experimental translation using cardinality constraints
 e.g.  X + Y -> A + B  becomes:
   EquivalentTo: (has_participant_collection some ((has_component exactly 1 X) and (has_component exactly 1 Y)) 
   and has_participant_collection some ((has_component exactly 1 A) and (has_component exactly 1 B))
   
  One for the reasoner competition?  Scaling might be OK if we treat MF axiomatisation as a separate module.

Consolidation of group calls into one (on USC agenda)

Intent to merge all the calls into one? LEGO, Annotation, PAINT, and Ontology Development calls into one GO call.

 Yes, that was the intent. Concerns that merging the ontology call with the rest is too much (different audiences, different topics from annotations related calls) - to be discussed at the USC meeting.