add CTD gene-disease (direct) associations

[nlwashington]

We need the CTD gene-disease associations:

from http://ctdbase.org/downloads/CTD_genes_diseases.tsv.gz

these are relatively straightforward. we only want to pull the direct (not-inferred) associations. each row becomes a G2PAssociation. you'll need to follow the pattern of creating a BNode of "some variant of Gene X" to link to the disease. you can see how that was done in OMIM. (we can consider abstracting this as a method to the G2PAssociation.)

Field structure is: GeneSymbol, GeneID (NCBI Gene identifier), DiseaseName, DiseaseID (MeSH or OMIM identifier), DirectEvidence ('|'-delimited list), InferenceChemicalName, InferenceScore, OmimIDs ('|'-delimited list), PubMedIDs ('|'-delimited list)

Luckily, they use NCBIGene ids, so that's easy.

I would say that IF there is a only one OMIM id for the DiseaseID or in the OmimIDs list, use that preferentially over any MeSH id. @cmungall will comment on if we should include or exclude annotations to MeSH for the first pass due to MONDO restrictions.

[nlwashington]

these have the following as "evidence". but it's really the relationship:

• marker/mechanism: A gene that may be a biomarker of a disease (e.g., increased expression of gene X correlates with breast cancer) or play a role in the etiology of a disease (e.g., mutations in gene X causes liver cancer).
• therapeutic: A gene that is or may be a therapeutic target in the treatment a disease (e.g., targeted reduction of gene X expression reduces susceptibility to emphysema).

so, there's actually different kinds of associations to make here; not just "has_phenotype".

so, because we don't know if it is causal, the ones for "marker/mechanism" should probably just have the relationship MONARCH:correlates_with (until @cmungall adds a correlation relationship to RO, and we'll switch out to the proper RO)

and for therapeutic, that's more complicated. let's also leave those out for now...these can be added after we finish the modeling related to variant -> "activity of gene X" -> disease.

[bryanlaraway]

@nlwashington On your last comment, how about for direct evidence = 'marker/mechanism|therapeutic.' Leave out or include with the MONARCH:correlates_with?

[bryanlaraway]

This is just about wrapped up, but testing the output. The disease2gene file has 40,148,312 rows. It will complete when setting a limit, but haven't had it complete on the full file. Might have to just push it to the Bamboo server and let it run there. Running locally on my laptop, it processes about 555 rows per second, which puts the run time at just over 20 hours. Not sure if it will just die when it goes to write the turtle output, as I wouldn't be surprised if writing all the nodes created from these rows overloads the memory.

[nlwashington]

the disease processing should finish... we are only taking those that are direct associations, so there are way less than 40M rows of those. I'll take a look to see if we can speed that up.

and yes, we'd probably use that generic correlates_with for those that are the combo "evidence" because we just don't know what it actually is!

[nlwashington]

@cmungall do we have a final verdict on the proper relationship to use for something like "correlates with"?

[nlwashington]

ok, so the general pattern here so far is to use an OMIM id for a disease-gene association when an OMIM id is available (and to use it preferentially even if we have a MeSH id, if the MeSH:OMIM is 1:1). since we don't know if the association is actually a marker or a (etiological mechanism), we leave this as some kind of correlative relationship. if there is no OMIM id, then we can use the MeSH id. (many of the cancers are annotated to MeSH, and more atomic-phenotypes.) however there are some oddities in MeSH. For example:

• PSEN1 is a marker/mechanism for Disease Models, Animal
• SOD3 is a marker/mechanism for Disease Progression
• KCNJ8 is a marker/mechanism for Genetic Predisposition to Disease

As with OMIM, we create an anonymous variant locus, and link that to the disease/phenotype.

[nlwashington]

we will use the is_marker_for relationship here for the "marker/mechanism" relationship.

[nlwashington]

[nlwashington]

suggestions for possible scrubbing, as well as pushing feedback to the source: Dog Diseases MESH:D004283 Disease Models, Animal MESH:D004195 Genetic Diseases, Inborn MESH:D030342 Genetic Diseases, X-Linked MESH:D040181 Genetic Predisposition to Disease MESH:D020022

[nlwashington]

I've written CTD personnel to ask clarification about the odd "diseases".

[nlwashington]

i've temporarily added scrubbing of the above identifiers in commit d82f2d7...no associations will be added to our turtle files for those five ids.

[nlwashington]

@mbrush please review cmaps and close if satisfied.