[Demo] Versioned artifact demo: create protein design with 3 iterations done coding:7 reasoning:6

[Demo] Versioned artifact demo: create protein design with 3 iterations

Goal

TREM2 Context

Three Versions

v1: Initial AlphaFold Prediction

{
  "artifact_type": "protein_design",
  "title": "TREM2 Ectodomain Variant — AlphaFold Base",
  "version_number": 1,
  "version_tag": "initial",
  "metadata": {
    "uniprot_id": "Q9NZC2",
    "method": "alphafold",
    "sequence": "MEPLR...(truncated)...RDHPG",
    "mutations": [],
    "predicted_stability": 0.72,
    "binding_affinity": {"target": "Aβ oligomers", "kd_nm": 850},
    "plddt_score": 0.89,
    "design_rationale": "Wild-type ectodomain structure as baseline for optimization"
  }
}

v2: Stability-Optimized

{
  "version_number": 2,
  "version_tag": "stability-opt",
  "changelog": "Introduced 3 stabilizing mutations from Rosetta energy minimization",
  "metadata": {
    "method": "rosetta_relax",
    "mutations": ["T96S", "K186R", "A192V"],
    "predicted_stability": 0.91,
    "binding_affinity": {"target": "Aβ oligomers", "kd_nm": 780},
    "delta_G_kcal": -2.3,
    "design_rationale": "Rosetta-guided mutations to improve thermostability without disrupting binding interface"
  }
}

v3: Binding-Affinity Tuned

{
  "version_number": 3,
  "version_tag": "binding-opt",
  "changelog": "Interface mutations to enhance Aβ binding while maintaining stability gains",
  "metadata": {
    "method": "rosetta_interface_design",
    "mutations": ["T96S", "K186R", "A192V", "H157Y", "R62W"],
    "predicted_stability": 0.88,
    "binding_affinity": {"target": "Aβ oligomers", "kd_nm": 120},
    "delta_G_kcal": -1.8,
    "design_rationale": "Interface residue optimization for enhanced Ab oligomer recognition. Slight stability trade-off accepted for 6.5x binding improvement."
  }
}

Demo Walkthrough

Acceptance Criteria

Dependencies

• a17-18-VERS0001 (version tracking schema)
• a17-19-TYPE0001 (protein_design artifact type)
• a17-20-VAPI0001 (create_version, get_history, diff APIs)

Work Log

2026-04-03 22:08 PT — Slot 20

• Started task: Create versioned TREM2 protein design demo
• Read spec and verified dependencies (a17-18-VERS0001, a17-19-TYPE0001, a17-20-VAPI0001) are all complete
• Examined artifact_registry.py: register_artifact, create_artifact_version, get_version_history, diff_versions functions exist
• Created register_trem2_direct.py to register 3 versions via direct SQL (to avoid DB lock issues)
• Successfully registered all 3 versions:

• Linked to 5 existing TREM2 hypotheses
• Updated api.py:artifact_detail() to display version history timeline with:

• Ready to test after API restart

2026-04-04 05:09 PT — Slot 10

• Reacquired task lease after stale-task reap and validated spec scope against Quest 16 demo priorities.
• Reworked register_trem2_direct.py into an idempotent demo loader with deterministic IDs (protein_design-trem2-demo-v1..v3), explicit version lineage, and derives_from links.
• Added hypothesis-linking logic to connect all 3 versions to top 5 existing TREM2 hypothesis artifacts through artifact_links.
• Executed timeout 300 python3 register_trem2_direct.py successfully: created all three versions and 15 hypothesis support links.
• Verified lineage via SQLite (version_number 1→2→3, parent chain intact, only v3 is_latest=1).
• Verified API outputs:

• Verified page rendering:

• Ran timeout 300 python3 link_checker.py (required timeout wrapper): completed with pre-existing large broken-link inventory; results written to logs/link-check-latest.json.

2026-04-14 02:55 PT — Slot 54

• Critical bug fix: get_version_history() was walking backward only (following parent_version_id chains), which correctly found ancestors but failed to find descendants. For TREM2 v1 (which has no parent, but has children v2→v3), the function returned only [v1] instead of [v1, v2, v3].
• Fix applied: Rewrote get_version_history() in scidex/atlas/artifact_registry.py to:

• Verified: Starting from any of the 3 TREM2 versions, get_version_history() now returns all 3 versions in order (v1, v2, v3)
• Also verified: diff_versions(v1, v3) correctly shows accumulated mutations (0→3→5) and binding affinity improvement (850nM→120nM, 7x)
• Committed: [Atlas] Fix get_version_history to walk forward through children [task:d16-20-AVER0001] → 68a5e80f1
• Pushed: git push origin HEAD --force-with-lease to orchestra/task/ce1f232d-push-main-race
• Root cause: The original implementation assumed version history meant "ancestors only" (parent chain). But SciDEX uses child references (parent_version_id on the child pointing to parent) for forward lineage. The function was reversed in direction — it could find ancestors but not descendants.