Skills: DisMech

DisMech Skill

Browse and query the DisMech disease mechanism knowledge graph — 750+ curated disorders with pathophysiology mechanisms, gene associations, clinical phenotypes, and therapeutic targets.

Purpose

The dismech skill provides access to the DisMech knowledge base, a curated collection of disease mechanism entries. Each entry is authored by Claude using a structured YAML format with PubMed evidence citations and HPO/HGNC/MAXO ontology bindings, then validated by human curators via a PR-based workflow.

Note that this is a preliminary collaboration with the Monarch team, partially to support the use of TypeDB as a knowledge representation framework. Exploring the use of Skillful Alhazen for this work should be considered an unencumbered exploration of the domain and does not intrude upon or impact the excellent, original work being done on DisMech within their system.

This skill lets the agent query DisMech for disease-specific information: what genes are involved, what phenotypes are associated, what pathophysiology mechanisms are described, and what therapeutic targets exist.

Prerequisites

• Docker running (TypeDB starts automatically via make build-db)
• The DisMech database populated (run dismech.py ingest after cloning the YAML data)

Commands

uv run python .claude/skills/dismech/dismech.py <command> [args]

Command	What it does
ingest	Load DisMech YAML entries into the TypeDB dismech database
list-diseases	List all diseases in the database
show-disease	Display full disease entry including mechanisms, phenotypes, genes
search	Full-text search across disease entries
query-phenotypes	Find diseases associated with given HPO terms
query-genes	Find diseases associated with given genes (HGNC)
stats	Show database statistics (disease count, mechanism count, etc.)
add-evidence	Add a PubMed evidence citation to an existing disease entry

Database Isolation

DisMech runs in its own TypeDB database (dismech), separate from alhazen_notebook. This means:

• Disease entities in DisMech cannot be directly linked to entities in alhazen_notebook (no shared schema root)
• PubMed evidence in DisMech cannot be cross-referenced to scilit-paper entities ingested via scientific-literature
• The aboutness relation (for attaching notes to entities) is not defined in the DisMech schema

These are tracked as open schema gaps. See sciknow-io/alhazen-skill-dismech issues for current status.

Typical Workflow

You: Show me the DisMech entry for Marfan syndrome
You: What genes are associated with connective tissue disorders in DisMech?
You: Search for diseases involving the RAS/MAPK pathway
You: Show me all diseases with the HPO phenotype "aortic root dilation"

DisMech Curation Pipeline

DisMech entries are authored using a YAML format with LinkML-validated structure. Each entry requires:

• PubMed PMIDs with exact verbatim quotes from abstracts as evidence
• HPO terms for phenotypes
• HGNC gene symbols
• MAXO treatment terms

Submissions go through a GitHub PR workflow with automated CI review (claude-code-review.yml) that checks PMID authenticity, quote accuracy, and ontology binding correctness before human approval.

To contribute a new disease entry, see the DisMech repository for the YAML schema and submission instructions.

Source Repository

The dismech skill lives in its own repository: github.com/sciknow-io/alhazen-skill-dismech. Gap issues for this skill are filed there, not in the main alhazen-skill-examples repository.