Ontology meeting 2015-11-05
Attendees: Melanie, Tanya, Paola, David H, David OS, Heiko, Harold, Chris
Propagating taxon constraints from dependent ontologies
Necessary for JBMS paper. Where are we with this please: https://github.com/geneontology/go-ontology/issues/12128
Fiddly problem. Two possible solutions are sketched in GH ticket above.
AI: Chris and Heiko to discuss this today, and look into implementing asap so we have an idea of how many annotations would be flagged as incorrect. Chris to then update the GH ticket.
In general, we should go for stricter taxon rules where these are implemented in e.g. UBERON where experts have given thought about them. We can always relax later down the line if new evidences emerge.
DNA binding terms
Should 'single-strand DNA binding' and 'double-stranded DNA binding' really sit under 'structure-specific DNA binding'? This is defined as "Interacting selectively and non-covalently with DNA of a specific structure or configuration e.g. triplex DNA binding or bent DNA binding.". Any DNA molecule is either single-stranded or double-stranded (except for triplexes), so where's the 'specificity' if it all falls under 'structure-specific DNA binding'?
AI: In the short term, Paola to place ‘single-stranded DNA binding’ and ‘double-stranded DNA binding’ directly under ‘DNA binding’. Then look at other descendants of ‘DNA binding’ and place them as most appropriate under ‘single-stranded DNA binding’, ‘double-stranded DNA binding’, ‘sequence-specific DNA binding’ where such links may be necessary and are currently missing. At the end, evaluate if ‘structure-specific DNA binding’ may be reworded based on its remaining children, and whether it is indeed a necessary term. As of today there are ~170 direct manual annotations, will need to examine them. All of this has been recorded in the GH ticket above, so it is not forgotten.
Response to compounds used as assays
Stemming from TG requests for
cellular response to doxorubicin [PMID:19801496]
response to tetrachloromethane [PMID:7852267]
cellular response to tetrachloromethane [PMID:7852267]
I emailed the submitter (and the group) but there was no reply: "Doxorubicin and tetrachloromethane are artificial compounds, not found physiologically in cells. In the papers above (and in many others) they are used to elicit a specific effect - doxorubicin intercalates DNA; tetrachloromethane is powerfully toxic to the liver. My view is that these substances are used as assays, and as such, curators should not annotate to corresponding 'response to' terms. We discussed the 'assay' issue in general at a couple of recent GOC meetings (Texas and Barcelona, you may search the minutes with 'assay' e.g. here https://docs.google.com/document/d/1NonH97s8xEpDdx6DfonKPKI_RdHbbG-yft85UZtUmF0/edit#), and I think that your requested terms fall under the same category. Therefore, I'd like to obsolete them.
Note that we do have 'response to doxorubicin' and 'response to 4'-epidoxorubicin' in GO. However, these were requested by the Experimental Factor Ontology for mapping purposes. They use those terms to annotate experiments, not gene products, and as such, we believe that their presence in the GO is legitimate.
I realize that this may be confusing. Do you [Stan and editors] think curators would benefit from a reminder on annotation guidance in this case? If yes, we could spend 5' on this at an annotation call. Should we place 'response to doxorubicin', 'response to 4'-epidoxorubicin' and any similar one in the 'do not annotate' subset?
AI: In the short term, Paola to obsolete the TG FF terms above (and email Stan to point him to this discussion). DONE.
AI: In the medium term, Paola to place 'response to doxorubicin', 'response to 4'-epidoxorubicin' and any similar one in the 'do not annotate' subset. (For example, search for terms requested by Dani Welter and Helen Parkinson.) Other editors to do the same when they come across terms of any kind (response to, binding…) that have a logical axiom to a non-physiological chemical compound. All to add definition comments meant as annotation guidance (e.g. for 'response to doxorubicin' something along the lines of “This term is not meant to be used to annotate gene products. It was created to be used for cross-referencing by other ontologies” in addition to the usual warning for ‘do not annotate’ terms.).
AI: In the long term, when we start using or referencing an assay ontology, we may evaluate alternative strategies with ontologies such as EFO. For example, they could annotate to the relevant chemical or drug by using a relation ‘exposure_to’ or similar.
Adding logical defs to cellular response terms
DOS has started to do this. What's the syntax, and should we look into filling all gaps.
Decision: All agree that "occurs_in some cell" is sufficient for cellular response to X terms. While this pattern wil not work for all 'cellular' terms, it is likely to be applicable in multiple branches.
AI: DOS to add 'response to' patterns to GitHub design pattern repo
Follow-up on Inference management
Documentation of ETINE flagging & relationship to redundancy stripping.
STILL TO DO: Document how ETINE flagging works :Chris:Heiko:
AI: DOS to try to track down examples of potential bugs found during previous discission of ETINE results.
Problem: How to break down assesment of existing ETINEs? Need to work through all of these before moving to automated deletion. There are over 1500.
Follow-up on Design Patterns
Need to work on automating DP validation via Travis. Requires running Jython on Travis while pulling in relevant libs.
AI: DOS and Heiko to meet next week & screenshare to work through how to do this.