Ontology meeting 2014-04-10

From GO Wiki
Jump to navigation Jump to search

Attendees: Jane, Paola, David H, David OS, Heiko, Tanya, Chris

Minutes: Paola

Follow-up: Abstract for Bio-Ontologies SIG at ISMB

Any comments on the semi-final text? Deadline for edits is Friday. Google doc is here:

https://docs.google.com/document/d/1KFCa7LDvz8bEBC_gFmzpt_xPM6KyZXh3cSiqSwT806U/edit

Resolution: Jane emailed Judy with our thoughts.

OBO to OWL roundtrip broken by never_in_taxon

For good reasons, never_in_taxon relationships are converted to annotation axioms in OWL. But this breaks OBO<->OWL roundtripping - required for any editing of GO axioms in Protege.

e.g.

   relationship: never_in_taxon NCBITaxon:33090 {id="GOTAX:0010013"} ! Viridiplantae
   >
   property_value: never_in_taxon NCBITaxon:33090 {id="GOTAX:0010013"}

Possible fixes:

  1. keep never_in_taxon in OBO as property_value: never_in_taxon NCBITaxon:33090 {id="GOTAX:0010013"}. Edit in Protege only.
  2. Add a new ad hoc rule to oboformat owl2obo translation that convert never_in_taxon annotation property axiom back to a relationship.

Resolution: in the short term, we'll go with option 1. (In OE, we'll still be able to see 'only_in_taxon'.)

For the longer term scenario, Chris and David O-S will discuss.

Issues with adding starts_with, ends_with links - has_part, really

Stems from trying to add starts_with, ends_with links to 'apoptotic process' (summary of discussion between Paola and David H):

"A programmed cell death process which *begins* when a cell receives an internal (e.g. DNA damage) or external signal (e.g. an extracellular death ligand), and *proceeds* through a series of biochemical events (signaling pathways) which typically lead to rounding-up of the cell, retraction of pseudopodes, reduction of cellular volume (pyknosis), chromatin condensation, nuclear fragmentation (karyorrhexis), plasma membrane blebbing and fragmentation of the cell into apoptotic bodies [these are all bits of the execution phase]. The process *ends* when the cell has died. The process is divided into a signaling pathway phase, and an execution phase, which is triggered by the former."

First issue: 'death' as the end point we define there is not a process; more of a phenotype.

Second issue: the links should then be: 'apoptotic process' starts_with 'apoptotic signaling pathway', 'apoptotic process' ends_with 'execution phase of apoptosis'. But starts_with and ends_with are instances of has_part, so we'd have a cycle error, because currently the signaling and execution phases are part_of 'apoptotic process'. If we remove the part_of links, we'll lose annotations because they don't propagate over has_part. This was noted before. The fudge of making dual annotations (to e.g. both 'apoptotic process' and 'apoptotic signaling pathway') is risky because a) it inflates the number of annotations in areas covered by has_part links (usually focus of more recent ontology development) and b) annotation policy should be made consistent - all DBs should adopt the dual annotation policy. I remember that Chris has suggested to use a different relationship alternative to has_part that would allow to keep the part_of links, but I can't find the email where this was discussed. In any case, we probably should put a hold on adding starts_with, ends_with links until this is resolved.

Update: Paola added the starts_with, ends with link without removing the part_ofs. As expected, this generated additional warnings in OE. Then turned off all cycle checks in OE as we get valid checks from Jenkins anyway. All seems fine so far. Also, edited the definition to say that "When the execution phase is completed, the cell has died."

Resolution: keep adding both part_of and has_part links where appropriate. All agree with changes in apoptotic defs.

Adding vesicle terms

See emails from Andrew Su and Kei Cheung. We'll see what their list looks like and then either give them a TG FF login, or add the terms ourselves.

Waiting for them to email us a list. No further action is required from our part for now.

Causal relationships

Should we be adding relationships between processes where one is causally dependent on another. e.g.

learned vocal behaviour causally_dependent_on some vocal learning

(See discussion here: https://sourceforge.net/p/geneontology/ontology-requests/10565/#2e93 )

Seems reasonable in cases like this, but I suspect that there are lots of hidden gotchas in trying to apply more generally - and a whole branch of philosophy with different solutions. Perhaps a big can of worms we should avoid unless we can come up with pressing use cases?

Resolution: David O-S will use causally_upstream_of for this case.

part_of nucleus (or other top-level CC term) vs is_a nuclear part

Stems from this SF request from Val (see last comments): https://sourceforge.net/p/geneontology/ontology-requests/10757/

Val argues that community curators drilling down from 'nucleus' would find it odd and confusing to see granular terms such as 'nuclear rDNA heterochromatin'. Issues:

1) We now have a TG template for part_of_cell_component; 'nuclear rDNA heterochromatin' was created via that template (is_a rDNA heterochromatin, part_of nucleus)

2) Our guidelines encourage curators to add part_of links (http://beta.geneontology.org/page/maintaining-complete-isa-and-partof-trees-cellular-component). They do mention 'is_a x part' links but we don't draw the line anywhere. E.g. Birgit adds part_of nucleus/membrane links to her protein complex terms.

Resolution: We'll keep the part_of links, then the is_a links will be automatically inferred.

For Val: we can work towards putting all the x-part terms in a separate subset; the browsers don't need to display them, though we're not sure if the community annotation tool can do this. Chris will put the anti-slim in. He'll come up with a convention first.

Update: Chris commented on https://sourceforge.net/p/geneontology/ontology-requests/10757/: "There's no need for most browsers or annotation tools to show the X-part terms. I put these into an antislim (goantislim_grouping, unless we can figure a better name), and they may eventually go away altogether".

How difficult would it be to get OBO-EDIT to save an OWL rather than an OBO file?

Filters wouldn't work, so a giant file would be saved. Plus we'd rather not do it because the diffs would get really ugly.

Pharma slim

Jane reported on a recent meeting she had at EBI with some people from Roche. They developed their own kind of slim that don't use GO, ~360 slim terms; then they tried to map it to GO terms. They'd want things such as 'aminoacid metabolism' and 'aminoacid transport' grouped together, so that they could do searches on e.g. "all processes that involve chemical x (e.g. angiotensin) or cell type x"; i.e. a slim with union classes. We discussed this and concluded that we'll produce a slim for them, but we wouldn't really want to do it that way. David H points out that in the long term transport and metabolism may be linked. David OS suggests, Could we prepare some canned queries in AmiGO2 to retrieve terms/annotations say occurring in cell-type x? So people could use them. Chris reminds us that we don't have enough logical defs pointing to anatomical terms. We could just give them a mapping between their terms and ours, while we wait for Uberon-based logical defs to go into production. Or we'll automate it in some way as it might be useful to us too. David OS will try to come up with sample queries; or Jane will look into the commonalities among terms (we'll likely go for the latter option).

Rest of JIRA issues (bumped to next week)

Try to review those we didn't get to last week.