Review response to drug hierarchy for GO vs EFO terms

[siiraa]

As a general rule, if a drug is available in Chebi, the corresponding 'response to' term should be created in GO and imported into EFO. Review all subclasses of response to drug that currently have an EFO ID to see if an appropriate GO term could be generated to replace them.

[paolaroncaglia]

A few thoughts about this:

• [ ] Check with GO if it's still in scope for them to create 'response to drug' terms with addition of appropriate comment (see e.g. https://www.ebi.ac.uk/ols/ontologies/go/terms?iri=http%3A%2F%2Fpurl.obolibrary.org%2Fobo%2FGO_1902522). AFAIK such terms would need to be added manually (but using design patterns). Or are we ok with keeping, and continuing to add, EFO terms under 'response to drug'? If so, should we revise (add to) their axiomatisation similarly to GO, e.g. 'response to 4'-epidoxorubicin' has_input CHEBI:47898 '4'-epidoxorubicin'? (https://www.ebi.ac.uk/QuickGO/term/GO:1902522)
• [ ] The EFO hierarchy under 'response to drug' is quite flat at the moment. Depending on the strategy we choose (see question above), the flatness may be resolved by aligning with GO (#444) and/or CHEBI. But some or all EFO-only terms may need to be rehoused manually (e.g. EFO:0007880 'response to dendritic cell-based immunotherapy', based on its definition, should be a child of EFO:0004645 'response to vaccine').
• [ ] EFO:0009166 'response to ivacaftor - efficacy' "The effectiveness of response to the drug, Ivacaftor, used to treat cystic fibrosis patients with mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene." should not be under 'response to drug', maybe modify label to make that explicit?

[paolaroncaglia]

This ticket is currently in our Icebox, but when we come back to it, we may want to consider how Mondo is approaching representation of responses to drugs (https://github.com/monarch-initiative/mondo/issues/860).

[paolaroncaglia]

Update: the Mondo ticket has now been addressed, so I'll move this to Backlog, and when we have a chance, we could examine Mondo's approach and evaluate adopting it for EFO too.

[paolaroncaglia]

Updating this ticket to document recent discussion (on the GO tracker) re. obsoletion of 'response to drug' GO terms, some of which are used in EFO by GWAS curators: https://github.com/geneontology/go-ontology/issues/21998

[kallia-p]

@paolaroncaglia @zoependlington @matentzn @eks-ebi @dosumis Update: We are reviewing the representation and axiomatisation of all 'response to' terms currently in EFO. These comprise terms that were requested and imported from GO and terms that were generated in EFO as per the GWAS catalog and OTAR needs. The GO superclass is "response to stimulus" GO:0050896 with Subclasses "response to chemical" GO:0042221 of which "response to drug" GO:0042493 is a Subclass Of. We aim to align between EFO, OBA, GO, VT and CHEBI. EFO 'response to' terms that aren't in GO will be added to OBA. This is in line with MONDO 's approach: obsoleted all response to terms and added them in OBA (monarch-initiative/mondo#860). We will do a templated addition to OBA + xref/SSSOM mapping from EFO to OBA.

Template: https://docs.google.com/spreadsheets/d/13qh7dLE38vMyz91oRqj6GzKjFohNazNKAJxKG6Plw1o/edit#gid=213599470

Task 1: Manual mappings |   |   -- | -- | --   | 1 | Search on OLS (https://www.ebi.ac.uk/ols/ontologies/oba) whether suitable OBA term   | 2 | if suitable OBA term add OBA id in oba_id column, and OBA label in object_label)   | 3 | if not suitable OBA term, search OLS in VT (https://www.ebi.ac.uk/ols/ontologies/vt)   | 3a | if suitable VT term add VT id in vt_id column, and VT label in object_label)   | 4 | if no suitable OBA or VT term, create new OBA id in general_response table   | 4a | Search on GO on OLS (https://www.ebi.ac.uk/ols/ontologies/go) whether suitable response term   | 4b | if suitable GO response term, add to go_class column   |   | if no suitable GO response term, add GO:0042221 (response to chemical) or GO:0050896 (response to stimulus) to go_class column   | 5 | Look for a suitable CHEBI term on OLS (https://www.ebi.ac.uk/ols/ontologies/chebi)   | 5a | If CHEBI term, add to "chemical_entity" column   | 5b | If no CHEBI term, check PRO term on OLS (https://www.ebi.ac.uk/ols/ontologies/pr)   | 5c | If PRO term, add to chemical_entity column   | 6 | If chemical_entity is a drug used for treaments, write "yes" in the "drug" column We will also consider adding broad 'response to' subclasses in order to differentiate between 'toxicity/adverse effects' and 'response to treatment with' in line with the GWAS cat, OTAR and ClinVar (see monarch-initiative/mondo#860). OTAR requests top level terms terms such as drug resistance, drug toxicity which they wanted to be in the biological mechanism branch https://github.com/ebispot/efo/issues/1209

[kallia-p]

@dosumis @zoependlington @matentzn Work in this ticket should also address https://github.com/EBISPOT/efo/issues/166 We should probably be stripping is_about links to CHEBI compounds as part of work on this branch. ('has input' is the correct relation - based on GO schema).

[kallia-p]

@matentzn @dosumis

Just an update on the manual mapping approach and the slight modification to the instructions that Nico and I discussed. Briefly we discussed adding a couple of columns for the GO nearest parent and its label. This is because (for the terms that weren't in OBA or in VT) I was originally asked to add either response to stimuli or response to chemical under go_class; but in many instances the nearest parent in GO is more specific e.g. GO:0097334 - response to perphenazine has a nearest parent in GO response to antipsychotic drug. Therefore for this term the general go_class_label is response to chemical but the nearest_go_parent_label is response to antipsychotic drug.

For some other terms in GO there is a mutli-hierarchy of nearest parents so to reflect I have added all nearest parents separated by "|". e.g. GO:0097335 - response to quetiapine has two parental classes: response to organonitrogen compound | response to organic cyclic compound

In some instances and usually for EFO terms the nearest parents differ between GO and EFO (which I presume is because there was an attempt in EFO to introduce some structure which is not mirrored in GO) . e.g. EFO:0010103 - response to peginterferon alfa-2a has a nearest parent response to interferon in EFO. This term is not represented in GO but if it was it would be under nearest parent 'response to drug'. therefore for these type of terms I am still adding go_class 'response to chemical' and go_nearest parent 'response to drug' and I am documenting the EFO parents in the Comment box.

Nearest parents can also differ between EFO and GO for terms that are the same e.g. EFO:0007592 - response to bleomycin GO:1904975 - response to bleomycin. Nearest parent in EFO is GO:0097327 (response to antineoplastic agent) whereas nearest parent in GO is GO:1901652 (response to peptide)

[helenp]

Some of the response to chemical terms are so old that I was the creator of them. I think the nearest parents are an indicator of divergence between GO and EFO over time.

[kallia-p]

Thanks, yes indeed, for the last example above GO added response to peptide a year ago.

[d0choa]

Please don't constrain the modelling based on our previous requests (e.g. #1209). We are quite keen to get this information organised in EFO and we will adapt to whatever conclusion you reach.

As mentioned before, we have several datasets with response or resistance to drugs very often in the context of a disease: GWAS catalogue, ClinVar, Cancer Genome Interpreter, etc. Our biggest concern is to find an approach that scales with the amount of data generated.

[kallia-p]

@rays22 see comments for response to terms