Define how to include evidence from model organism studies

[caufieldjh]

we should have a better way to include model organism recapitulaton and not have it be part of the primary evidence for phenotype in human

Originally posted by @cmungall in #61 (comment)

[github-actions]

Issue Summary: Model Organism Evidence Integration

Context

This issue emerged from PR #61 (Holt-Oram Syndrome curation) where @cmungall noted that model organism data (e.g., mouse arrhythmia phenotypes) was being mixed with primary human phenotype evidence without clear distinction. The concern: we need better separation between human disease evidence and supporting model organism recapitulation data.

Current State

The dismech schema already has an AnimalModel class (lines 1330-1340 in schema) with the following structure:

AnimalModel:
  slots:
    - species
    - genotype
    - background
    - genes
    - category
    - alleles
    - description
    - associated_phenotypes
    - evidence

The Disease class includes animal_models as a top-level slot alongside phenotypes, pathophysiology, etc. (line 1312).

The Problem

However, the current usage pattern allows model organism evidence to leak into human phenotype descriptions:

1. Phenotype.evidence accepts any EvidenceItem with a PMID reference
2. There's no validation to distinguish human vs. model organism studies
3. Curators may cite mouse/rat studies as direct evidence for human phenotypes
4. This conflates translational support with direct clinical observation

Proposed Solutions

Option 1: Evidence Source Typing (Minimal Schema Change)

Add an optional evidence_source field to EvidenceItem:

EvidenceItem:
  slots:
    - reference
    - supports
    - snippet
    - explanation
    - evidence_source  # NEW: human_clinical | model_organism | in_vitro | computational

EvidenceSourceEnum:
  permissible_values:
    HUMAN_CLINICAL:
      description: Direct observation in human patients
    MODEL_ORGANISM:
      description: Evidence from animal models (mouse, rat, zebrafish, etc.)
    IN_VITRO:
      description: Cell culture or in vitro experiments
    COMPUTATIONAL:
      description: In silico predictions or models

Pros: Minimal disruption, backward compatible, allows filtering in queries
Cons: Relies on curator discipline, doesn't structurally separate the evidence

Option 2: Separate Evidence Collections (Moderate Change)

Create distinct evidence slots for different source types:

Phenotype:
  slots:
    - human_evidence      # Evidence from human studies only
    - model_evidence      # Links to AnimalModel entries
    - in_vitro_evidence   # Cell-based evidence

Pros: Clear structural separation, enforces best practices
Cons: More invasive schema change, requires migration of existing data

Option 3: Bidirectional Linking (Current + Enhancement)

Keep the current animal_models top-level section, but enhance it:

AnimalModel:
  slots:
    - species
    - genotype
    - recapitulated_phenotypes  # NEW: List of human phenotypes this model recapitulates
    - evidence
    
Phenotype:
  slots:
    - evidence               # Human-only evidence
    - model_support          # NEW: References to AnimalModel entries by name/ID

Pros: Maintains current structure, makes relationships explicit
Cons: Requires curator awareness of the linking pattern

Recommendation

I recommend Option 1 (Evidence Source Typing) as the most pragmatic first step:

1. Add evidence_source enum to EvidenceItem class
2. Update curation SOP in CLAUDE.md to require source classification
3. Add validation rule: Warn if Phenotype.evidence cites a model organism PMID without evidence_source: MODEL_ORGANISM
4. Enhance HTML rendering to visually distinguish evidence types (e.g., badges/icons for human vs. model)

Later, we can consider Option 3 for richer semantic linking between animal_models and phenotypes.

Related Work

• Alliance of Genome Resources: Separates human disease associations from model organism phenotype annotations
• Monarch Initiative: Uses distinct evidence/association types for cross-species phenotype matching
• ClinGen: Classifies evidence by provenance (patient data, functional studies, model organisms)

Next Steps

1. Decide on approach (Option 1, 2, 3, or alternative)
2. Update schema with new slots/enums
3. Update CLAUDE.md SOP with evidence sourcing guidelines
4. Migrate existing data if needed (audit current evidence for model organism citations)
5. Update HTML templates to render evidence source distinctions

Questions for Discussion

• Should we make evidence_source required or optional?
• Do we need finer-grained categories (e.g., separate zebrafish from mouse)?
• Should the AnimalModel.associated_phenotypes field reference human Phenotype entries?
• How do we handle studies that span both human and model organism data?

---

Labels: enhancement, documentation (already applied)