Investigate use of all trait names vs. preferred trait names in OT evidence generation

[apriltuesday]

Context of issue:
When we do trait mapping (automated and manual), we use only preferred names, but when we annotate we attempt to use all names. Because we retain previous mappings even if they don't appear (i.e. don't appear among preferred names in current ClinVar), this means obsolete mappings can be not just retained but also used without being updated.

Example - in ClinVar:

    <TraitSet Type="Disease" ID="6307">
      <Trait ID="4675" Type="Disease">
        <Name>
          <ElementValue Type="Preferred">Malignant tumor of urinary bladder</ElementValue>
          <XRef ID="Bladder+cancer/7822" DB="Genetic Alliance"/>
          <XRef ID="399326009" DB="SNOMED CT"/>
        </Name>
        <Name>
          <ElementValue Type="Alternate">Urinary bladder cancer</ElementValue>
          <XRef ID="MONDO:0001187" DB="MONDO"/>
        </Name>
        <Name>
          <ElementValue Type="Alternate">Urinary Bladder Neoplasms</ElementValue>
          <XRef ID="D001749" DB="MeSH"/>
        </Name>
        <Name>
          <ElementValue Type="Alternate">Bladder cancer</ElementValue>
        </Name>
        <AttributeSet>
          <Attribute Type="keyword">Hereditary cancer syndrome</Attribute>
        </AttributeSet>
        <XRef ID="MONDO:0001187" DB="MONDO"/>
        <XRef ID="C0005684" DB="MedGen"/>
        <XRef Type="MIM" ID="109800" DB="OMIM"/>
      </Trait>
    </TraitSet>

In latest mappings:

# preferred name yields up-to-date mapping
$ grep -i '^Malignant tumor of urinary bladder' latest_mappings.tsv
malignant tumor of urinary bladder	http://purl.obolibrary.org/obo/MONDO_0004986	urinary bladder carcinoma

# alternate name yields obsolete mapping
$ grep -i '^Urinary bladder cancer' latest_mappings.tsv
urinary bladder cancer	http://www.ebi.ac.uk/efo/EFO_0000292	bladder carcinoma

In #383 we modified annotated XML generation to use only preferred names, observing that it decreased coverage of traits only slightly while decreasing the number of obsolete EFO terms used significantly.

The goal of this issue is to see what is the impact of making a similar change for OT evidence string generation (which is more complicated due to how it groups and explodes traits), and if it is acceptable make the change.

[apriltuesday]

See also #210

[apriltuesday]

This issue has resurfaced in the course of investigating how to clean up our trait mappings (#428). I think it's important that we consistently perform the trait mapping and the evidence generation using the same set of trait names, whether it's the preferred names or all names, so we should discuss what makes the most sense for Open Targets.

From looking at past issues like #189 and opentargets/issues#1013, it looks like this behaviour was agreed upon under the assumption that the multiple trait names within a single ClinVar trait would map to the same EFO term. It's not clear that this is the case.

Here's another example, this time excerpted from the 2025.12 evidence strings:

# Snippets of 3 evidence strings for the same RCV
{..."cohortPhenotypes": ["BITEMPORAL FORCEPS MARKS SYNDROME", "FFDD type 2", "FOCAL FACIAL DERMAL DYSPLASIA, TYPE II", "Focal facial dermal dysplasia 3", "Focal facial dermal dysplasia 3, Setleis type", "Focal facial dermal dysplasia type III", "SETLEIS SYNDROME"],
    "diseaseFromSource": "Focal facial dermal dysplasia type III",
    "diseaseFromSourceMappedId": "Orphanet_1807", ...}
{..."cohortPhenotypes": ["BITEMPORAL FORCEPS MARKS SYNDROME", "FFDD type 2", "FOCAL FACIAL DERMAL DYSPLASIA, TYPE II", "Focal facial dermal dysplasia 3", "Focal facial dermal dysplasia 3, Setleis type", "Focal facial dermal dysplasia type III", "SETLEIS SYNDROME"],
    "diseaseFromSource": "Focal facial dermal dysplasia type III", 
    "diseaseFromSourceMappedId": "Orphanet_398166", ...}
{..."cohortPhenotypes": ["BITEMPORAL FORCEPS MARKS SYNDROME", "FFDD type 2", "FOCAL FACIAL DERMAL DYSPLASIA, TYPE II", "Focal facial dermal dysplasia 3", "Focal facial dermal dysplasia 3, Setleis type", "Focal facial dermal dysplasia type III", "SETLEIS SYNDROME"],
    "diseaseFromSource": "Focal facial dermal dysplasia type III",
    "diseaseFromSourceMappedId": "MONDO_0009203", ...}

These three EFO IDs are actually mapped to various strings within cohortPhenotypes (i.e. all trait names within ClinVar), even though only one of those strings is singled out in diseaseFromSource (i.e. the preferred trait name within ClinVar, or lexicographically first name if no preferred name is given). It's not clear to me whether all the terms in that list should really map to the same ID (e.g. FOCAL FACIAL DERMAL DYSPLASIA, TYPE II and Focal facial dermal dysplasia type III). This would suggest that we should be using all trait names, in order to capture all the associations potentially implied by the ClinVar record.

On the other hand, presumably ClinVar has grouped these terms under a single trait for a reason, and we should do the same - namely take the preferred trait name and the EFO ID mapped to that one, maybe only falling back to other names if the preferred name is not yet mapped. According to the above comment from 2023, this would only decrease coverage of the traits (i.e. how many have EFO mapping) by a small amount, while being IMHO clearer and more predictable than the current strategy.

@tcezard @ireneisdoomed @DSuveges @vivienho Let me know what you think, or we can discuss at our next meeting.