Structural Biology & Disease Mechanisms Project

Overview

Diseases where protein structure determination (PDB X-ray crystallography, cryo-EM) or computational structure prediction (AlphaFold) has been key to understanding pathophysiology, enabling drug design, or interpreting variant pathogenicity.

This is both a curation guide (which diseases to prioritize) and a schema design exploration (how structural biology insights fit into dismech's data model).

How Structural Biology Data Fits Into dismech

Where it already fits (no schema changes needed)

1. Evidence items — PDB/AlphaFold papers are citeable evidence for mechanism claims: ```yaml pathophysiology:

   • name: BCR-ABL constitutive kinase activation evidence:
     • reference: PMID:11423618 # imatinib-ABL crystal structure paper snippet: "Crystal structure of ABL in complex with STI571..." evidence_source: COMPUTATIONAL ```

2. Treatment.mechanism — Structure-based drug design narratives: ```yaml treatments:

   • name: Imatinib mechanism: >- Binds the inactive conformation of ABL kinase domain. Structure-based design informed by PDB:1IEP crystal structure. ```

3. ComputationalModel — AlphaFold or structural models as computational tools: ```yaml computational_models:

   • name: AlphaFold2 structure of CFTR ΔF508 model_type: MACHINE_LEARNING # or a new STRUCTURAL_PREDICTION type description: >- AlphaFold2 prediction of CFTR with ΔF508 deletion reveals destabilization of NBD1-ICL4 interface. repository_url: https://alphafold.ebi.ac.uk/entry/P13569 ```

4. GeneticContext.functional_impact — Variant-level structural effects: ```yaml genetic_basis:

   • allele_type: missense functional_impact: >- T315I gatekeeper mutation blocks imatinib binding by introducing steric clash at the ATP-binding pocket. ```

Where the schema could grow (potential new features)

Option A: Add structural_insights to ComputationalModelTypeEnum

Minimal change — just add a new enum value:

ComputationalModelTypeEnum:
  permissible_values:
    STRUCTURAL_PREDICTION:
      description: >-
        Protein structure prediction (AlphaFold, RoseTTAFold) or
        molecular dynamics simulation of disease-relevant proteins
    MOLECULAR_DOCKING:
      description: >-
        Computational docking of drug candidates to protein targets,
        typically informed by PDB structures

This lets structural biology data live in the existing computational_models slot with proper typing. Least disruptive.

Option B: Add structural_context slot to GeneticContext

For variant-level structural mapping:

structural_context:
  description: >-
    Structural biology context for how mutations affect protein
    structure or drug binding. References PDB/AlphaFold entries.
  range: StructuralContext
  inlined: true

# New class
StructuralContext:
  slots:
    - pdb_id
    - alphafold_id
    - affected_domain
    - structural_effect   # e.g., "disrupts salt bridge", "steric clash"
    - drug_binding_impact  # e.g., "blocks imatinib binding pocket"
    - evidence

More structured but heavier — only worth it if we plan to systematically capture variant-structure mappings.

Option C: Add protein_structures top-level slot on Disease

Like computational_models but specifically for structural biology:

protein_structures:
  - name: ABL kinase domain with imatinib
    pdb_id: PDB:1IEP
    resolution: 2.1 A
    method: X-RAY_CRYSTALLOGRAPHY
    disease_relevance: >-
      Defines the imatinib binding mode and explains resistance mutations
    key_residues:
      - T315 (gatekeeper)
      - E286 (DFG motif)
    evidence: [...]

Most expressive but adds significant schema surface area.

Recommendation

Start with Option A (add enum values to ComputationalModelTypeEnum) — it's zero-breaking-change and lets us capture structural biology in existing computational_models blocks immediately. As we curate, we'll learn whether Options B or C are needed. The functional_impact free-text field on GeneticContext already handles variant-structure narratives adequately for now.

Diseases to Curate / Annotate with Structural Biology

Tier 1: Structure Directly Enabled Targeted Therapy

These are paradigmatic — structure-based drug design changed patient outcomes.

Tier 2: Structure Revealed Disease Mechanism

Understanding pathophysiology, even if therapy didn't follow directly from structure.

Tier 3: AlphaFold-Era Breakthroughs

Where AlphaFold specifically opened new doors.

Implementation Plan

Phase 1: Schema (minimal)

• [ ] Add STRUCTURAL_PREDICTION and MOLECULAR_DOCKING to ComputationalModelTypeEnum
• [ ] Validate enum additions pass just qc

Phase 2: Annotate Existing Entries

• [ ] ATTR Amyloidosis — add TTR structure / tafamidis binding to computational_models
• [ ] ALK NSCLC — add ALK kinase structures to computational_models
• [ ] Alzheimer's — add cryo-EM fibril structures
• [ ] Achondroplasia — add FGFR3 structural context
• [ ] Phenylketonuria — add PAH structure mapping
• [ ] Fanconi Anemia — enrich cryo-EM references
• [ ] HIV/AIDS — add protease/RT/integrase structures

Phase 3: New High-Impact Entries

• [ ] CML (BCR-ABL + imatinib — the poster child)
• [ ] Cystic Fibrosis (CFTR modulator story)
• [ ] Sickle Cell Disease (first molecular disease)
• [ ] HER2+ Breast Cancer (trastuzumab structural basis)

Phase 4: Evaluate Schema Needs

After Phase 2-3 curation, assess whether: - functional_impact free text is sufficient for variant-structure narratives - A dedicated StructuralContext class (Option B) would add value - A top-level protein_structures slot (Option C) is warranted

References & Resources

• PDB: https://www.rcsb.org/
• AlphaFold DB: https://alphafold.ebi.ac.uk/
• PDBe Knowledge Base: https://www.ebi.ac.uk/pdbe/pdbe-kb/
• SIFTS (Structure Integration with Function, Taxonomy, Sequence): maps PDB to UniProt
• Missense3D: predicts structural impact of missense variants
• AlphaMissense: DeepMind pathogenicity classifier using AlphaFold