Skip to content

feat: add rare disease MDT copilot team with ontology-first workflow#99

Closed
hazelian0619 wants to merge 1 commit intoaristoteleo:mainfrom
hazelian0619:feature/rare-disease-team
Closed

feat: add rare disease MDT copilot team with ontology-first workflow#99
hazelian0619 wants to merge 1 commit intoaristoteleo:mainfrom
hazelian0619:feature/rare-disease-team

Conversation

@hazelian0619
Copy link
Copy Markdown

Summary

Introduce a production rare disease multi-agent team for evidence-backed differential reasoning. The "Rare Disease MDT Copilot" coordinates six specialized agents through a structured clinical decision-support workflow.

What's included

Agent definitions (6 agents)

  • leader — workflow coordinator, case structuring, delegation, candidate synthesis
  • phenotype_structurer — free-text → structured HPO-aligned phenotype package
  • evidence_researcher — ontology-first evidence retrieval (rd_ontology → BioMCP → web)
  • genotype_analyst — variant/gene interpretation when genomic data is available
  • auditor — contradiction and evidence-quality review before final delivery
  • reporter — clinician-facing structured report writer

Team template

  • rare_disease_team — assembles the 6 agents with a hard workflow contract

Ontology layer

  • rd_ontology ToolSet — agent-native SQLite queries against Orphanet/OMIM/HPO (6 tools: resolve_term, search_disease, get_disease, get_hpo_term, find_by_hpo, stats)
  • rd_ontology_first skill — enforces ontology-before-online retrieval pattern
  • ETL pipeline (build_rd_ontology.py) — Orphanet XML + HPO .obo/.hpoa + OMIM → rd_ontology.sqlite
  • CLI query helper (query_rd_ontology.py) — resolve disease aliases, fetch detailed records

Architecture

Dual-layer design: offline ontology layer (Orphanet/OMIM/HPO → local SQLite) for deterministic normalization, then online evidence layer (BioMCP/Web/DB API) for literature and current evidence.

Design notes

  • The team is designed for clinician support, not automatic diagnosis
  • All agents follow a strict ontology-first workflow: standardize before search
  • The skill is a single consolidated file loaded by 3 agents (leader, phenotype_structurer, evidence_researcher)
  • Non-goals explicitly documented in each agent: no definitive diagnosis claims, no treatment decisions, uncertainty must be preserved

🤖 Generated with Claude Code

Introduce a production rare disease multi-agent team for evidence-backed
differential reasoning. The team coordinates six specialized agents
through a structured workflow: phenotype normalization → evidence
retrieval → genotype analysis → audit → clinician-facing report.

Core additions:
- 6 agent definitions (leader, phenotype_structurer, evidence_researcher,
  genotype_analyst, auditor, reporter)
- rare_disease_team template assembling the full workflow
- rd_ontology ToolSet providing agent-native SQLite queries against
  Orphanet/OMIM/HPO data
- rd_ontology_first skill enforcing ontology-before-online retrieval
- ETL pipeline (build_rd_ontology.py) and CLI query helper
  (query_rd_ontology.py)

Architecture follows a dual-layer design: offline ontology layer
(Orphanet/OMIM/HPO → local SQLite) for deterministic normalization,
then online evidence layer (BioMCP/Web/DB API) for literature.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
@hazelian0619
Copy link
Copy Markdown
Author

Superseded by v2.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant