GeneDiseaseAssociation predicates

[joeflack4]

Overview

Right now, we are using the predicate RO:0003303 (_causes condition_) to represent all gene::disease associations (i.e. all rows in morbidmap.txt). I don't think this was our intention; I just think this is where Dazhi left it about a year ago and I'm just realizing this now.

Courses of action

Sabrina's has a google doc with a great collection of predicates we can consider for this: https://docs.google.com/spreadsheets/d/1bzxGT7vqQNhUHhe3vcOv-5JDGZg5izS5D7J2M2zYBxs/edit#gid=1090676757

Nearly every entry in morbidmap.txt has 1 mapping key. I believe there are rare edge cases of no mapping key, and I found 1 instance where there are 2 mapping keys (#81). The mapping keys come with descriptions (found in a comment at the bottom of morbidmap.txt) which we can use to help us determine better predicates:

# 1 - The disorder is placed on the map based on its association with a gene, but the underlying defect is not known. # 2 - The disorder has been placed on the map by linkage or other statistical method; no mutation has been found. # 3 - The molecular basis for the disorder is known; a mutation has been found in the gene. # 4 - A contiguous gene deletion or duplication syndrome, multiple genes are deleted or duplicated causing the phenotype.

Additional info

Here's how we're currently representing gene::disease associations in the latest release (the structure of this will be greatly changed very soon).

OMIM:100678 a owl:Class ;
    rdfs:label "ACAT2" ;
    RO:0002525 CHR:9606chr6q25.3 ;
    ...
    RO:0003303 OMIM:614055 ;
    biolink:category biolink:Gene ;
    biolink:has_evidence "The disorder is placed on the map based on its association with a gene, but the underlying defect is not known." .

Some example rows from morbidmap.txt (I put one row for each of the 4 mapping keys).

Phenotype   Gene Symbols    MIM Number  Cyto Location
?ACAT2 deficiency, 614055 (1)   ACAT2   100678  6q25.3
?Anal canal carcinoma (2)   ANC 105580  11q22-qter
17,20-lyase deficiency, isolated, 202110 (3)    CYP17A1, CYP17, P450C17 609300  10q24.32
?Pain sensitivity QTL1 (4)  PAINQTL1    618377  1p33

Related

• 77

[joeflack4]

@sabrinatoro I saw that in #75 you mentioned that:

The curly brackets means this is a susceptibility term.

Example:

Phenotype   Gene Symbols    MIM Number  Cyto Location
{Type 2 diabetes mellitus, susceptibility to}, 125853 (3)   GPD2    138430  2q24.1

Just wanting to add note about that here. I'm assuming that one of the predicates in your RO spreadsheet might be appropriate for susceptibility.

[sabrinatoro]

I think that each mapping key should be associated with a specific relation as they mean something different:

1 - The disorder is placed on the map based on its association with a gene, but the underlying defect is not known.

I don't think these represent actual diseases. All the examples I checked represent locus and are excluded. I think we can ignore.

2 - The disorder has been placed on the map by linkage or other statistical method; no mutation has been found.

This means that we know that a gene is associated with a disease, but we don't know what variation/mutation is responsible for this disease I suggest we use: RO_0003303 (causes condition) note: gene2disease relation, might need a reciprocal one for disease2gene !!! see the caveat below regarding multiple genes.

3 - The molecular basis for the disorder is known; a mutation has been found in the gene.

This means that we know that gene is associated with a disease AND we know what variation/mutation is responsible. for OMIM, I suggest we use: RO_0004013 (is causal germline mutation in) note: gene2disease relation, might need a reciprocal one for disease2gene !!! see the caveat below regarding multiple genes.

4 - A contiguous gene deletion or duplication syndrome, multiple genes are deleted or duplicated causing the phenotype.

This means that a disease was associated with multiple genes because the variation associated with the disease is a huge deletion/duplication involving multiple genes, and we don't know which one is the one causal one. I suggest we use: RO_0003304 (contributes to condition)

!!!!!!!!!!!!!!Important caveats

• Some entry in the "gene symbols" column represent variation, and not actual genes. For example: 46XY sex reversal 4 (4) "DEL9p24.3, C9DELp24.3, SRXY4" 154230 9p24.3 DEL9p24.3, C9DELp24.3 are variations SRXY4 is a gene

• The relations for 2 and 3 are only for the cases when there is only 1 gene associated with the disease. If there is more than one gene associated with the disease, the suggested relation to use is either RO_0004016 (is causal germline mutation partially giving rise to) if we KNOW that the disease is digenic/olygogenic, or RO_0003304 (contributes to condition)

• "?", before the phenotype name indicates that the relationship between the phenotype and gene is provisional. In these cases, we should probably skip and not curate the gene-disease relation

We should review this as a group and come into an agreement before doing anything. I will review some examples.

[joeflack4]

This is really awesome, and even more than I had hoped for. You mentioned reviewing before doing anything. I'm already so far in with adding these predicates (along with other changes in this pr) that I would hate to go back to using just plain RO_0003303 for everything like we were doing before.

Until I come back from Thanksgiving, I'm going to leave that PR in a state where it simply uses these predicates for each of the mapping keys (per your defaults above):

MORBIDMAP_PHENOTYPE_MAPPING_KEY_PREDICATES = {
    '1': None,  # these will be skipped
    '2': RO['0003303'],
    '3': RO['0004013'],
    '4': RO['0003304'],
}

I realize there's a ton of nuance and more complexity here, so I know I may have to change this before my PR is merged, or later otherwise.