| name | toxicology-semantic-model |
| description | Semantic model for the toxicology database (mutagenesis molecular data — molecules, atoms, bonds, connectivity). Load this skill when the user is querying the toxicology database. |
| metadata | {"author":"manual","version":"1.1","last_updated":"2026-02-25T00:00:00Z","learned_patterns":0,"verified_patterns":4} |
Toxicology — Semantic Model
Available Resources
| Resource | What it contains | How to load |
|---|
| Extended gotchas | Label encoding, atom ID format, graph traversal pitfalls | read_skill_resource("toxicology-semantic-model", "references/gotchas.md") |
| Metric definitions | Atom counts, toxicity prevalence, bond type distributions | read_skill_resource("toxicology-semantic-model", "references/metrics.md") |
| Query patterns | Canonical SQL for toxic molecules, element queries, graph traversal | read_skill_resource("toxicology-semantic-model", "references/query_patterns.md") |
When to load resources:
- If querying for toxic/non-toxic molecules → load
references/gotchas.md (label encoding is critical)
- If traversing the molecular graph (paths, connectivity) → load
references/gotchas.md + references/query_patterns.md
- If counting atoms, bonds, or computing distributions → load
references/metrics.md
- For simple lookups by molecule ID → the information below is sufficient
Key Concepts
| Term (user says) | Actual meaning | Confidence |
|---|
| "toxic" / "mutagenic" / "positive" | molecule.label = '+' — string plus sign, NOT integer 1 | verified |
| "non-toxic" / "non-mutagenic" / "negative" | molecule.label = '-' — string minus sign, NOT integer 0 | verified |
| "single bond" | bond.bond_type = '-' | verified |
| "double bond" | bond.bond_type = '=' | verified |
| "triple bond" | bond.bond_type = '#' | verified |
| "aromatic bond" | bond.bond_type = ':' | verified |
| "carbon atom" | atom.element = 'c' (lowercase) | verified |
| "connected atoms" | connected table — links atom pairs via atom_id, atom_id2, bond_id | verified |
Gotchas
-
molecule.label is a string, not an integer — toxic molecules have
label = '+', non-toxic have label = '-'. Writing WHERE label = 1
will silently return zero rows.
-
atom.element values are lowercase — carbon is 'c', not 'C' or
'Carbon'. Chlorine is 'cl', nitrogen is 'n', oxygen is 'o'.
-
Atom and bond IDs encode structure — 'TR000_1' means molecule TR000,
atom position 1. Bond ID 'TR000_1_2' is the bond between atoms 1 and 2
in TR000. Do not treat these as opaque strings when debugging.
-
Graph traversal requires recursive CTEs — single-hop joins on connected
only find immediate neighbours. For path-length queries, use WITH RECURSIVE.
-
bond and connected serve different purposes — bond stores bond type
per molecule; connected maps which atom pairs share a bond. Both are needed
for element-and-bond-type queries.
For extended gotchas and SQL examples, call:
read_skill_resource(
skill_name="toxicology-semantic-model",
resource_name="references/gotchas.md"
)
Core Join Paths
molecule.molecule_id ──→ atom.molecule_id (molecule ↔ its atoms)
molecule.molecule_id ──→ bond.molecule_id (molecule ↔ its bonds)
atom.atom_id ──→ connected.atom_id (atom ↔ connections, one end)
atom.atom_id ──→ connected.atom_id2 (atom ↔ connections, other end)
bond.bond_id ──→ connected.bond_id (bond ↔ atom pair sharing it)
There is no direct FK between atom and bond; go through connected to
link a specific atom to the bonds it participates in.
Frequently Asked Queries
For canonical query patterns (toxic molecules by element, bond type filters,
graph traversal), call:
read_skill_resource(
skill_name="toxicology-semantic-model",
resource_name="references/query_patterns.md"
)
Metric Definitions
For full metric formulas (atom counts, toxicity rates, bond distributions), call:
read_skill_resource(
skill_name="toxicology-semantic-model",
resource_name="references/metrics.md"
)
