pyowl2-profiles

Fast OWL 2 profile detection (EL/RL/QL/DL) over pyoxigraph.

Installation

pip install pyowl2-profiles

Quick start

import pyoxigraph
from pyowl2_profiles import detect_profiles, PROFILE_NAMES

# Load your ontology into a pyoxigraph store
store = pyoxigraph.Store()
with open("my-ontology.ttl", "rb") as f:
    store.load(f.read(), pyoxigraph.RdfFormat.TURTLE)

# Run detection
result = detect_profiles(store, graph_iri=None)

# Check verdicts
for profile in PROFILE_NAMES:  # ("el", "rl", "ql", "dl")
    r = result[profile]
    status = "IN" if r["in_profile"] else "OUT"
    print(f"OWL 2 {profile.upper()}: {status}  ({r['total_violations']} violations)")

# Inspect violation samples
for sample in result["el"]["sample_violations"][:3]:
    print(f"  {sample['axiom_type']}: {sample['subject_iri']}")

Why this library

• No JVM. ROBOT requires Java; pyowl2-profiles runs entirely in Python with no external runtime.
• No second parse. If you already have data in a pyoxigraph store (e.g., from a pipeline), detection adds only a few lightweight SPARQL queries.
• Fast. On moderate-sized ontologies (< 1M triples), detection completes in under a second. Cold-start (load + detect) is 10-1400x faster than ROBOT depending on ontology size.
• Standard Python packaging. pip install, import, done.

CLI usage

Detect profiles for a file

pyowl2-profiles detect my-ontology.ttl
pyowl2-profiles detect my-ontology.ttl --format json
pyowl2-profiles detect my-ontology.ttl --profile EL

Benchmark against ROBOT

# Requires robot on PATH
pyowl2-profiles benchmark onto1.ttl onto2.owl --csv results.csv

# Skip ROBOT, ours only
pyowl2-profiles benchmark onto1.ttl --skip-robot

Validation

Validated against ROBOT 1.9.10 on a 19-ontology biomedical fleet: 19/19 (100%) verdict agreement across all four profiles (EL, RL, QL, DL).

See docs/usage.md for a detailed walkthrough and benchmark methodology.

API reference

detect_profiles(store, graph_iri=None, ontology_id="", version_id="") -> dict

Run all four OWL 2 profile checks against a pyoxigraph.Store.

Parameters:

• store — pyoxigraph.Store containing the ontology triples
• graph_iri — named graph IRI to query, or None for the default graph
• ontology_id — optional string identifier (informational only)
• version_id — optional version string (informational only)

Returns a dict with keys "el", "rl", "ql", "dl", "indexed_at". Each profile value is a dict with:

• in_profile (bool) — True if no violations found
• total_violations (int) — total violation count
• violations_by_axiom_type (dict[str, int]) — per-type counts
• sample_violations (list[dict]) — up to 50 example violations, each with axiom_type, subject_iri, manchester (rendered tokens), and optionally details

ProfileViolation

Frozen dataclass for a single violation: profile, axiom_type, subject_iri, details, manchester.

PROFILE_NAMES

("el", "rl", "ql", "dl") — the four OWL 2 profiles in subset order.

ProfileName

Type alias: Literal["el", "rl", "ql", "dl"].

Acknowledgement

Originally extracted from OntoExplorer, a FAIR ontology repository at Maastricht University Institute of Data Science.

License

MIT — see LICENSE.