Ontology meeting 2016-01-21

From GO Wiki
Jump to: navigation, search

Attendees: DavidH, DavidOS, Harold, Heiko, Melanie, Paola, Tanya

Minutes: Tanya

Fun with regulates axioms


 400 some new inferences after Chris added some regulates axioms before the holiday break. After fixing what was causing the
 build to break, we can now examine them. The problem described in 
 https://github.com/geneontology/go-ontology/issues/12235 was one result that was troubling.  See this ticket.  
 Basic pattern is 'negative regulation of xx by negative regulation of yy' inferred SubClassOf: positive regulation of xx.  
 This was due to a badly designed TG template (since removed by Heiko). But the bigger problem is that this is just too 
 much complexity to pack into a term.  All agreed ee should get rid of the terms we have and not allow more. 
 There are 22 terms following this pattern.  We have decided to obsolete all that have no annotations ASAP.
  For those that have annotations,  as a temporary measure we will delete the Equivalence axioms 
  that cause the problem and manually assert classification.  We will then request re-annotation for all existing annotations, 
  preferably via construction of a LEGO model -> production of inferred annotations in GAF would give annotation to  the relevant 
  parent classes.
 Discussion of practicality of doing this in LEGO:
  When will be able to do this?  
  GAF from LEGO status:  Basic inference already works. This is probably sufficient for the models we need for this issue.
  But sometimes we want more than just the basic inference.  T
  The whole part  where we go to deeper terms, use inferencing features, is taking longer.  What do we state in the model and what do we infer?
  Needs discussion. Something stable is needed by the end of the month will be out for KMW to consume.  Less than 2 weeks to the
  basic version.

DNA methylation and gene expression

Stemming from latest assert report (the same one with the negative by negative terms)

 ADD	GO:1901535 'regulation of DNA demethylation' GO:0010468 'regulation of gene expression'

because 'DNA demethylation' (and 'DNA methylation') is_a 'regulation of gene expression, epigenetic'. Should it be part_of instead?

 DOS: I have a mini project going on this topic.  
   From discussion with Ruth and Karen, we agreed that a regulates relation 
 should apply between 'chromatin modification' and 'epigenetic regulation of gene expression'
 Long confusing umbrella ticket here:  https://github.com/geneontology/go-ontology/issues/11628
 All agree that we are uncomfortable with DNA methylation is_a 'regulation of gene expression, epigenetic'
 This is not intuitive to biologists.  All more comfortable with using part_of.
 DOS: But if all DNA methylation regulates gene expression then it would be valid to add 
 DNA methylation SubClassOf: regulates some 'gene expression' and this would => inference that 
 DNA methylation SubClassof 'regulation of gene expression'.  It would also be consistent with the 
 definition of 'regulation of gene expression' and with the fact the regulates is now transitive (so indirect regulation is 
  allowed and will be inferred in many cases.) How can we avoid this?
  Suggestion: If we use causally_upstream_of, for cases like this, we can keep the causal link but prevent the odd inference.
  Also serves as a way of making breaks in causal chains to prevent inference of regulation going to far.
  DH: Just realised this is moot: restriction enzyme system in bacteria, methylation causes inability for REs to cut - DNA methylation 
 is not always linked to 'regulation of gene expression, epigenetic' - Kills the part_of solution as this is not always true.
 PMID:6243774.  Also means we can't use regulates or causally upstream of. Solution is to just delete the offending axiom.
UPDATE 27/1/16 Done, deleted links: 
'DNA demethylation' is_a 'regulation of gene expression, epigenetic'
'DNA methylation' is_a 'regulation of gene expression, epigenetic'
'regulation of DNA demethylation' is_a 'regulation of gene expression'
'regulation of DNA methylation' is_a 'regulation of gene expression'

Separate topic: discussion of 'regulates o part_of.

 DH: Discomfort with regulates o part_of. Imagines that this will be disastrous in LEGO models. Where molecular functions
 that are part of the same signal transduction pathway may have a directly activates relation between them. We wouldn't want to
 infer that they regulate the pathway. 
 DOS: Strongly agree. Definitely a problem for ST pathways.
 Consult with Val?
 Poss solutions: 
 Use 'causally upstream of' as a solution? More general than 'regulates' and no inference over part_of.  
 But this won't work for direct -ve and +ve reg steps in ST
 DOS: suggest we remove the regulates o part_of axiom.  
 Still very useful as a heuristic for finding potential regulates relationships.

Latest assertions

The list in yesterday's assert report (email "[go-ontology] build-go-assert-inferences - Build # 173 - Fixed!") is really long, 402 added axioms presumably as a result of the synapse branch merge. Has anyone had a chance to review them all?

 DOS: Inferences from synapse branch merge reviewed.
 Many inferences also come from the change to regulation axiomatisation.
 I have reviewed these briefly.  Some problematic ones are flagged in
 AI: Everyone else look through report for odd references.

Logical definitions for 'response to chemical' terms

Logical defs for 'response to chemical' terms point to 'response to stimulus', while 'cellular response to chemical' points to 'cellular response to chemical stimulus':

id: GO:1903842

name: response to arsenite ion

intersection_of: GO:0050896 ! response to stimulus

intersection_of: has_input CHEBI:22633 ! arsenite ion

relationship: has_input CHEBI:22633 {is_inferred="true"} ! arsenite ion


id: GO:1903843

name: cellular response to arsenite ion

intersection_of: GO:0070887 ! cellular response to chemical stimulus

intersection_of: has_input CHEBI:22633 ! arsenite ion

relationship: has_input CHEBI:22633 {is_inferred="true"} ! arsenite ion

Was this intentional - shouldn't the former point more specifically to 'response to chemical'? (which is_a 'response to stimulus') Of course, the 'response to chemical' parentage is inferred anyway.

DOS RESPONSE: [note - DH suggested the same]

Non-redundant classification is correct:

Screen Shot 2016-01-21 at 09.55.00.png

This equivalence axioms takes care of the correct classification of 'response to arsenite ion' under 'response to chemical':

  • 'response to chemical' EquivalentTo 'response to stimulus' and ('has input' some 'chemical entity')

We don't yet have axiomatisation to deal with the 'cellular bit' - hence the more specific genus in this case. But if we agree that all cellular response occurs in some cell we could have:

  • 'cellular response to stimulus' EquivalentTo: 'response to stimulus' and occurs in some cell
  • 'cellular response to chemical stimulus' EquivalentTo: 'response to stimulus' and occurs in some cell and has_input some 'chemical entity'
  • 'cellular response to arsenite ion' EquivalentTo: 'response to stimulus' and occurs in some cell and has_input some 'arsenite ion'

The reasoner would then look after classification completely.

 AI: DOS will check with Val whether she's comfortable with 'occurs in' some cell. Then we'll run by Chris next week.

movement, transport, establishment of localization

See this document that DOS put together: https://docs.google.com/document/d/1mLuIbLGo9MApOVBeELzqaalKDcLG0xZthWnporGVkCg/edit?usp=sharing


https://github.com/geneontology/go-ontology/issues/9189 which brings up issues with 'establishment of localization' and 'transport'

DOS' work in this area should clear up that issue as well.

 Major problem:  the branch under organelle transport is a TPV.  
 By current def, transport is for 'substances'  not big things like organelles, vesicles.
 But biologists use term transport for many classes under here (esp for vesicular transport along axons)
 Poss solutions:
  1. Redefine transport? But then how does it differ from 'establishment of loc' ?  Merge?
  2. Use a different term in this branch (tracfficking?), and move it up from under transport.
  3. Keep the term transport, but move term under establishment of loc
 Melanie suggests using synonyms and using 'community of usage' as a property. "you can them extract specific slims for your users
 with their preferred labels" Might that be a solution?  
 DOS: Could be part of solution, but don't think there's a community that uses anything but transport for vesicles moving along axons.
 More discussion needed.