Proposal: Add Variant-centric Associations to Capture Phenopacket Context

[kevinschaper]

Summary

This proposal outlines additional association types that could be generated from phenopacket data to better capture the variant-level relationships and their context.

Current Associations

Currently we generate:

• CaseToPhenotypicFeatureAssociation (168,288 edges)
• CaseToDiseaseAssociation (8,207 edges)
• CaseToGeneAssociation (8,138 edges) - newly fixed

Proposed New Associations

1. CaseToVariantAssociation

• Subject: Case (phenopacket.store:PMID_xxx)
• Object: Variant ID
• Properties: zygosity, interpretation_status
• Impact: ~8,207 associations (all records have variants)

2. VariantToGeneAssociation

• Subject: Variant ID
• Object: Gene ID (HGNC:xxx)
• Properties: interpretation_status
• Impact: ~8,138 associations

3. VariantToDiseaseAssociation

• Subject: Variant ID
• Object: Disease ID (OMIM:xxx)
• Properties: interpretation_status (CAUSATIVE, CONTRIBUTING)
• Impact: ~8,207 associations

Capturing Phenopacket Context on Non-Case Associations

For associations that don't directly involve Case nodes (VariantToGene, VariantToDisease), we need to preserve provenance back to the source case/phenopacket.

Existing Biolink Context Qualifiers (pattern)

• disease_context_qualifier - "A context qualifier representing a disease or condition in which a relationship expressed in an association took place"
• anatomical_context_qualifier - for anatomical locations
• species_context_qualifier - for taxonomic species
• population_context_qualifier - for population context

Gap Identified

No case_context_qualifier or individual_context_qualifier exists in biolink-model.

Proposal Options

1. Propose new slot to biolink-model (Recommended)

   • Add case_context_qualifier: Optional[str]
   • Description: "A context qualifier representing a case or individual in which a relationship expressed in an association was observed"
   • Would allow: case_context_qualifier: "phenopacket.store:PMID_xxx_yyy"

2. Use generic qualifiers list (fallback)

   • Available on all associations
   • Less semantic but works today
   • qualifiers: ["case:phenopacket.store:PMID_xxx_yyy"]

Additional Qualifier Usage

• interpretation_status (CAUSATIVE, CONTRIBUTING) → Use as statement_qualifier or dedicated property
• zygosity → association property (exists on some variant associations)
• onset_qualifier → already implemented for phenotypes/diseases

Data Available

From 8,207 phenopacket records:

• 100% have variant data with interpretation status
• 99% of variants have associated gene IDs (HGNC)
• 100% have disease associations

Next Steps

1. Decide on approach for case_context_qualifier (propose to biolink-model or use fallback)
2. Implement CaseToVariantAssociation transformer
3. Implement VariantToGeneAssociation transformer
4. Implement VariantToDiseaseAssociation transformer
5. Update transform.yaml with new edge_properties