Ontology meeting 2015-01-27
Attendees: Paola, David H, David OS, Harold, Heiko, Judy, Tanya, Jane, Chris
NOTE: THIS IS THE LAST MEETING HELD ON A TUESDAY
We will resume our usual Thursday schedule next week.
Do we want David OS to continue to work on this given staffing constraints? Is AmiGO in shape to receive this traffic anyway?
AI for DOS: he’ll email Andrew Su (and possibly others), cc Chris and go from there.
Are they appropriate for the ontology? The only one we currently have is 'integral component of plasma membrane'.
AI for David H: they’re out of scope for GO, unless there is evidence that the ligand is necessary for the receptor to form a functional complex.
Action items from previous week(s)
Everyone please check: http://wiki.geneontology.org/index.php?title=Ontology_meeting_2015-01-13
Anything we need to revise or discuss today?
Ontology group look into creating intermediate templates for complexes: AI for David H and Harold: make Trello ticket.
Readability of term labels: AI for DOS: make ticket and link it up to Trello.
Protege 4 vs 5: Protege 5 keeps crashing. AI: DOS will look into it.
Adding disjoints under 'binding' (and more broadly in the ontology)
Stemming from email thread "build-go-assert-inferences - Build # 117 - Successful!" (Bad asserted inference: ADD GO:0003680 'AT DNA binding' GO:0005515 'protein binding', stemming from 'AT DNA binding' has_input PR:000016450 ! transmembrane protease serine 11D)
[Chris] We should be thinking of ways to structure things and add disjoints to catch these [errors] further upstream, such as making the main grouping classes under classes like 'binding' disjoint (it doesn't have to exhaustive, that would be too hard)
E.g. if this subset of the hierarchy was PD:
binding macromolecule binding nucleic acid binding protein binding small molecule binding
[David H] This should work:
Macromolecule binding <disjoint with> small molecule binding
Nucleic acid binding <disjoint with> protein binding
The personally unsatisfying thing is all the other children of the binding term that are not going to fall into disjoint classes. We might want to have a look at those and see if we can tidy up the whole mess. Unlike the metabolism terms, the binding terms can follow on from disjoint relations of the chemical structures themselves.
Shall we make this part of a bigger (Trello) project to get disjoints into the ontology? Let's discuss and see what priority it should get.
Ideally, we’d want an automated solution: disjoints should live in ChEBI and we’d inherit them. May not work for metabolism, and it’s a big project, but if we really want ‘binding’ to work, we should work with ChEBI on this. ChEBI is struggling under the weight of their own work, so they may not be able to do it soon, but it appears that they’ve started to add disjoints already:
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3531142/ "Another extension of ChEBI, introduced as a downloadable OWL file in the same directory, is the assertion of disjoint relationships between high-level entities such as between ‘molecular entity’ and ‘group’." "A further change within the ontology has been the partitioning of all entities originally classified as carbohydrates into one of two disjoint classes, ‘carbohydrate’ and ‘carbohydrate derivative’."
(ftp://ftp.ebi.ac.uk/pub/databases/chebi/ontology/, disjoints last updated 18 months ago)
AI: We should ask ChEBI for a few more high-level disjoints, and work with them; it would help them too. Then we may look into automating if necessary/appropriate. Would require at least one of us to devote a significant chunk of time to this project.
[Paola, added after call] David OS and I will meet with Helen next week to discuss a GO-EBI strategic plan for 2015, before Helen meets the GO PIs in Hinxton. We may want to raise the topic with Helen then.
Add chain reasoning for regulation?
There are situations where we can infer a some
As we already assert that gene silencing subClassOf negative regulation of gene expression, this could be inferred by the property chain: -ve Reg o -veReg -> +veR
If this inference (and others like it) are always sound, we should add the property chains. But are they? In particular, do we need human judgment of the directness of regulation in order to safely make such inferences?
[David H] This should work. Nikolaj tried and we didn’t see any problem. People can’t just assume it’s direct regulation though.
AI: David H to present at the group, not next week though as he’ll be away. Also he’ll draft a definition of direct and indirect regulation.
AI: David OS will add the rules and send everyone the resulting report to look at. He’ll do that as a test in a separate branch. Heiko will help.
Remaining axioms in CHEBI that overclassify amino acids
What do do about glyco and lipid modifies AAs -that, chemically, are alpha amino acids.
AI: Harold to look into whether glyco and lipid modified AAs are ever found as free.
WE STOPPED HERE. ITEMS BELOW ARE BUMPED TO NEXT WEEK
Logical definition issues with partial membranes. Many membranes are classified as 'membrane part', but have no further classification or like 'dendrite membrane' have no indication that they are part of a dendrite.
We could get a lot more automated classification if we relax genus to something very general - probably 'cellular process'. There doesn't seem to be any over-inference that results - although we may need to stick to a convention of using has_output in the cellular_component biogenesis branch. Should we do this?
TermGenie templates for receptor activity
Please look at https://www.ebi.ac.uk/panda/jira/browse/GO-240
Where are we with this?
Need to review the situation as Chris wasn't on the call last week. (Paola has a list of tickets that need special attention. For my own reference, they're in an email thread "GO tickets in the EBI Jira instance".)
Obsoleted/merged terms in external ontologies we cross-ref to
Need solution. Reported in https://www.ebi.ac.uk/panda/jira/browse/GO-161