Dataset Viewer
Auto-converted to Parquet Duplicate
subject
stringlengths
8
279
predicate
stringlengths
9
166
object
stringlengths
1
1.3k
object_type
stringclasses
3 values
object_datatype
stringclasses
2 values
object_language
stringclasses
0 values
http://freebase.com//m/027rn
http://freebase.com//location/country/form_of_government
http://freebase.com//m/06cx9
uri
null
null
http://freebase.com//m/017dcd
http://freebase.com//tv/tv_program/regular_cast./tv/regular_tv_appearance/actor
http://freebase.com//m/06v8s0
uri
null
null
http://freebase.com//m/07s9rl0
http://freebase.com//media_common/netflix_genre/titles
http://freebase.com//m/0170z3
uri
null
null
http://freebase.com//m/01sl1q
http://freebase.com//award/award_winner/awards_won./award/award_honor/award_winner
http://freebase.com//m/044mz_
uri
null
null
http://freebase.com//m/0cnk2q
http://freebase.com//soccer/football_team/current_roster./sports/sports_team_roster/position
http://freebase.com//m/02nzb8
uri
null
null
http://freebase.com//m/04nrcg
http://freebase.com//soccer/football_team/current_roster./soccer/football_roster_position/position
http://freebase.com//m/02sdk9v
uri
null
null
http://freebase.com//m/07nznf
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/014lc_
uri
null
null
http://freebase.com//m/02qyp19
http://freebase.com//award/award_category/nominees./award/award_nomination/nominated_for
http://freebase.com//m/02d413
uri
null
null
http://freebase.com//m/0q9kd
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/0184jc
uri
null
null
http://freebase.com//m/03q5t
http://freebase.com//music/performance_role/regular_performances./music/group_membership/role
http://freebase.com//m/07y_7
uri
null
null
http://freebase.com//m/0gqng
http://freebase.com//award/award_category/winners./award/award_honor/ceremony
http://freebase.com//m/073hkh
uri
null
null
http://freebase.com//m/0b76d_m
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_distribution_medium
http://freebase.com//m/029j_
uri
null
null
http://freebase.com//m/014_x2
http://freebase.com//award/award_winning_work/awards_won./award/award_honor/award_winner
http://freebase.com//m/012ljv
uri
null
null
http://freebase.com//m/0ds35l9
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_region
http://freebase.com//m/05r4w
uri
null
null
http://freebase.com//m/015qsq
http://freebase.com//film/film/language
http://freebase.com//m/02bjrlw
uri
null
null
http://freebase.com//m/04bdxl
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/02s2ft
uri
null
null
http://freebase.com//m/09c7w0
http://freebase.com//location/location/contains
http://freebase.com//m/0rs6x
uri
null
null
http://freebase.com//m/079vf
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/0d90m
uri
null
null
http://freebase.com//m/015zyd
http://freebase.com//organization/organization/headquarters./location/mailing_address/country
http://freebase.com//m/09c7w0
uri
null
null
http://freebase.com//m/05vsxz
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/06qgvf
uri
null
null
http://freebase.com//m/04ljl_l
http://freebase.com//award/award_category/nominees./award/award_nomination/nominated_for
http://freebase.com//m/03qcfvw
uri
null
null
http://freebase.com//m/0grwj
http://freebase.com//people/person/profession
http://freebase.com//m/05sxg2
uri
null
null
http://freebase.com//m/0rh6k
http://freebase.com//location/statistical_region/religions./location/religion_percentage/religion
http://freebase.com//m/01lp8
uri
null
null
http://freebase.com//m/05d7rk
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award
http://freebase.com//m/027dtxw
uri
null
null
http://freebase.com//m/05hs4r
http://freebase.com//music/genre/artists
http://freebase.com//m/01pbxb
uri
null
null
http://freebase.com//m/03qcq
http://freebase.com//influence/influence_node/influenced_by
http://freebase.com//m/084w8
uri
null
null
http://freebase.com//m/01lxd4
http://freebase.com//music/genre/artists
http://freebase.com//m/0f0y8
uri
null
null
http://freebase.com//m/08815
http://freebase.com//education/educational_institution/students_graduates./education/education/student
http://freebase.com//m/05bnp0
uri
null
null
http://freebase.com//m/0g56t9t
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_region
http://freebase.com//m/05r4w
uri
null
null
http://freebase.com//m/0160w
http://freebase.com//organization/organization_member/member_of./organization/organization_membership/organization
http://freebase.com//m/02vk52z
uri
null
null
http://freebase.com//m/0dbpyd
http://freebase.com//people/person/place_of_birth
http://freebase.com//m/09c7w0
uri
null
null
http://freebase.com//m/053y0s
http://freebase.com//people/person/profession
http://freebase.com//m/0dz3r
uri
null
null
http://freebase.com//m/016qtt
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/09sh8k
http://freebase.com//film/film/other_crew./film/film_crew_gig/film_crew_role
http://freebase.com//m/09zzb8
uri
null
null
http://freebase.com//m/05pd94v
http://freebase.com//award/award_ceremony/awards_presented./award/award_honor/award_winner
http://freebase.com//m/01vvydl
uri
null
null
http://freebase.com//m/0gtsx8c
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_region
http://freebase.com//m/0jgd
uri
null
null
http://freebase.com//m/02r0csl
http://freebase.com//award/award_category/nominees./award/award_nomination/nominated_for
http://freebase.com//m/0m313
uri
null
null
http://freebase.com//m/02y_lrp
http://freebase.com//film/film/country
http://freebase.com//m/09c7w0
uri
null
null
http://freebase.com//m/012d40
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/034qmv
uri
null
null
http://freebase.com//m/07rlg
http://freebase.com//olympics/olympic_sport/athletes./olympics/olympic_athlete_affiliation/country
http://freebase.com//m/0b90_r
uri
null
null
http://freebase.com//m/0g22z
http://freebase.com//award/award_winning_work/awards_won./award/award_honor/award
http://freebase.com//m/05b4l5x
uri
null
null
http://freebase.com//m/018js4
http://freebase.com//film/film/country
http://freebase.com//m/09c7w0
uri
null
null
http://freebase.com//m/07fq1y
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/02qgqt
uri
null
null
http://freebase.com//m/028q6
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award
http://freebase.com//m/02581q
uri
null
null
http://freebase.com//m/0sxg4
http://freebase.com//film/film/genre
http://freebase.com//m/04xvlr
uri
null
null
http://freebase.com//m/0fvf9q
http://freebase.com//award/award_winner/awards_won./award/award_honor/award_winner
http://freebase.com//m/0l6qt
uri
null
null
http://freebase.com//m/0jz9f
http://freebase.com//film/film_distributor/films_distributed./film/film_film_distributor_relationship/film
http://freebase.com//m/083shs
uri
null
null
http://freebase.com//m/0njvn
http://freebase.com//location/statistical_region/rent50_2./measurement_unit/dated_money_value/currency
http://freebase.com//m/09nqf
uri
null
null
http://freebase.com//m/0d1tm
http://freebase.com//olympics/olympic_sport/athletes./olympics/olympic_athlete_affiliation/country
http://freebase.com//m/0154j
uri
null
null
http://freebase.com//m/01br2w
http://freebase.com//film/film/estimated_budget./measurement_unit/dated_money_value/currency
http://freebase.com//m/09nqf
uri
null
null
http://freebase.com//m/03rl1g
http://freebase.com//government/legislative_session/members./government/government_position_held/district_represented
http://freebase.com//m/05kkh
uri
null
null
http://freebase.com//m/0fbq2n
http://freebase.com//sports/sports_team/roster./american_football/football_historical_roster_position/position_s
http://freebase.com//m/01r3hr
uri
null
null
http://freebase.com//m/060bp
http://freebase.com//government/government_office_category/officeholders./government/government_position_held/jurisdiction_of_office
http://freebase.com//m/027nb
uri
null
null
http://freebase.com//m/04qvl7
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/nominated_for
http://freebase.com//m/02vxq9m
uri
null
null
http://freebase.com//m/04yywz
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/0140g4
uri
null
null
http://freebase.com//m/01k7d9
http://freebase.com//people/deceased_person/place_of_death
http://freebase.com//m/0k049
uri
null
null
http://freebase.com//m/02p65p
http://freebase.com//award/award_winner/awards_won./award/award_honor/award_winner
http://freebase.com//m/0337vz
uri
null
null
http://freebase.com//m/06688p
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/0b2v79
uri
null
null
http://freebase.com//m/01c0cc
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/01jc6q
http://freebase.com//award/award_winning_work/awards_won./award/award_honor/award
http://freebase.com//m/040njc
uri
null
null
http://freebase.com//m/0520r2x
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/0cb77r
uri
null
null
http://freebase.com//m/08815
http://freebase.com//education/educational_institution/students_graduates./education/education/student
http://freebase.com//m/06j0md
uri
null
null
http://freebase.com//m/01453
http://freebase.com//soccer/football_team/current_roster./sports/sports_team_roster/position
http://freebase.com//m/0dgrmp
uri
null
null
http://freebase.com//m/028_yv
http://freebase.com//film/film/country
http://freebase.com//m/03rjj
uri
null
null
http://freebase.com//m/01914
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/05_6_y
http://freebase.com//sports/pro_athlete/teams./sports/sports_team_roster/team
http://freebase.com//m/02b15h
uri
null
null
http://freebase.com//m/03c7ln
http://freebase.com//music/artist/track_contributions./music/track_contribution/role
http://freebase.com//m/026t6
uri
null
null
http://freebase.com//m/02pprs
http://freebase.com//music/instrument/instrumentalists
http://freebase.com//m/07s3vqk
uri
null
null
http://freebase.com//m/05zjtn4
http://freebase.com//education/educational_institution/students_graduates./education/education/student
http://freebase.com//m/0197tq
uri
null
null
http://freebase.com//m/02_fm2
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_distribution_medium
http://freebase.com//m/029j_
uri
null
null
http://freebase.com//m/0411q
http://freebase.com//people/person/profession
http://freebase.com//m/016z4k
uri
null
null
http://freebase.com//m/06151l
http://freebase.com//award/award_winner/awards_won./award/award_honor/award_winner
http://freebase.com//m/06gp3f
uri
null
null
http://freebase.com//m/01fq7
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/06w99h3
http://freebase.com//film/film/genre
http://freebase.com//m/02kdv5l
uri
null
null
http://freebase.com//m/05cljf
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/0342h
http://freebase.com//music/performance_role/regular_performances./music/group_membership/role
http://freebase.com//m/0l14qv
uri
null
null
http://freebase.com//m/02bfmn
http://freebase.com//people/person/gender
http://freebase.com//m/05zppz
uri
null
null
http://freebase.com//m/01rk91
http://freebase.com//business/job_title/people_with_this_title./business/employment_tenure/company
http://freebase.com//m/0kc6x
uri
null
null
http://freebase.com//m/01xdf5
http://freebase.com//base/popstra/celebrity/friendship./base/popstra/friendship/participant
http://freebase.com//m/04t2l2
uri
null
null
http://freebase.com//m/027qgy
http://freebase.com//film/film/genre
http://freebase.com//m/06cvj
uri
null
null
http://freebase.com//m/02v8kmz
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_distribution_medium
http://freebase.com//m/029j_
uri
null
null
http://freebase.com//m/01j5ts
http://freebase.com//people/person/place_of_birth
http://freebase.com//m/06_kh
uri
null
null
http://freebase.com//m/0dwl2
http://freebase.com//business/business_operation/industry
http://freebase.com//m/01mw1
uri
null
null
http://freebase.com//m/02wh75
http://freebase.com//award/award_category/winners./award/award_honor/award_winner
http://freebase.com//m/0lbj1
uri
null
null
http://freebase.com//m/0c3ybss
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_region
http://freebase.com//m/09c7w0
uri
null
null
http://freebase.com//m/01l1b90
http://freebase.com//common/topic/webpage./common/webpage/category
http://freebase.com//m/08mbj5d
uri
null
null
http://freebase.com//m/02vp1f_
http://freebase.com//film/film/release_date_s./film/film_regional_release_date/film_release_region
http://freebase.com//m/03_3d
uri
null
null
http://freebase.com//m/09m6kg
http://freebase.com//film/film/language
http://freebase.com//m/02h40lc
uri
null
null
http://freebase.com//m/05bp8g
http://freebase.com//people/person/nationality
http://freebase.com//m/03_3d
uri
null
null
http://freebase.com//m/047q2k1
http://freebase.com//film/film/genre
http://freebase.com//m/05p553
uri
null
null
http://freebase.com//m/01ypc
http://freebase.com//sports/sports_team/roster./baseball/baseball_roster_position/position
http://freebase.com//m/02dwn9
uri
null
null
http://freebase.com//m/06dv3
http://freebase.com//film/actor/film./film/performance/film
http://freebase.com//m/09m6kg
uri
null
null
http://freebase.com//m/01rtm4
http://freebase.com//education/educational_institution/students_graduates./education/education/student
http://freebase.com//m/02rchht
uri
null
null
http://freebase.com//m/0dq3c
http://freebase.com//business/job_title/people_with_this_title./business/employment_tenure/company
http://freebase.com//m/02zs4
uri
null
null
http://freebase.com//m/0qcr0
http://freebase.com//people/cause_of_death/people
http://freebase.com//m/0byfz
uri
null
null
http://freebase.com//m/06wzvr
http://freebase.com//film/film/genre
http://freebase.com//m/01q03
uri
null
null
http://freebase.com//m/01qn7n
http://freebase.com//tv/tv_program/languages
http://freebase.com//m/02h40lc
uri
null
null
http://freebase.com//m/0cnl80
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award_nominee
http://freebase.com//m/083chw
uri
null
null
http://freebase.com//m/0qf43
http://freebase.com//award/award_nominee/award_nominations./award/award_nomination/award
http://freebase.com//m/0f_nbyh
uri
null
null
http://freebase.com//m/0c0yh4
http://freebase.com//film/film/country
http://freebase.com//m/0d060g
uri
null
null
End of preview. Expand in Data Studio

🧬 Complete Ontology & Knowledge Graph Library v7

A curated, ready-to-use collection of ontologies, knowledge graphs (KGs), benchmarks, and cross-references β€” unified under a single 6-column Parquet schema optimized for LLM Agent tool-calling workflows.

All raw *.owl, *.ttl, *.obo sources are converted to standardized TSV / Parquet triple bundles with one Parquet shard per source (no sharding on small files). This library is designed to give Agents direct file paths, complete schemas, memory estimates, and runnable Python examples for every resource.


πŸ—ΊοΈ Repository Structure

dongdongkx/ontology/
β”œβ”€β”€ raw/                                    # 27 source ontologies (OWL/TTL/OBO)
β”‚   β”œβ”€β”€ gene-ontology/go.owl                # GO: ~71 K terms, 1.5 M axioms
β”‚   β”œβ”€β”€ go-plus/go-plus.owl                 # GO + relations merged (~2Γ— larger)
β”‚   β”œβ”€β”€ chemistry-entities-ontology/chebi.owl      # ChEBI: chemical structures
β”‚   β”œβ”€β”€ disease-ontology/doid.owl                   # DOID: disease IDs
β”‚   β”œβ”€β”€ mondra-ontology/mondo.owl                 # MONDO: unified disease ontology
β”‚   β”œβ”€β”€ uber-anatomy/uberon.owl                 # Uberon: comparative anatomy
β”‚   β”œβ”€β”€ cell-type/CL.owl                        # Cell type ontology
β”‚   β”œβ”€β”€ human-phenotype-ontology/hp.owl         # HP: clinical phenotype
β”‚   β”œβ”€β”€ experimental-factor/efo.owl             # EFO: experimental factors
β”‚   β”œβ”€β”€ sequence-ontology/so.owl                # SO: genomic features
β”‚   β”œβ”€β”€ organism-taxonomy/ncbitaxon.owl         # NCBI Taxonomy: 2.1 GB, 5 M+ triples ⚠️
β”‚   └── ...                                 # Full list in Section 1
β”‚
β”œβ”€β”€ kg/                                     # 10 KG datasets (Parquet triple shards)
β”‚   β”œβ”€β”€ yago-4.5/                           # 27.5 M rows, 6 shards β†’ General world facts
β”‚   β”œβ”€β”€ dbpedia-core-en/                    # 5 M rows, 8 shards β†’ English Wikipedia
β”‚   β”œβ”€β”€ dblp-bibliography/dblp.parquet       # DBLP: CS publication graph
β”‚   β”œβ”€β”€ conceptnet/conceptnet-5.8.parquet    # ConceptNet: commonsense relations
β”‚   β”œβ”€β”€ freebase-metaweb/freebase-metaweb.parquet # Freebase mirror (large)
β”‚   └── ...
β”‚
β”œβ”€β”€ benchmark/                               # 3 standard benchmarks (FB15K-237, WN18RR, NELL-995)
β”œβ”€β”€ ontology/                                # Pre-built mappings and serialized OWL files
β”œβ”€β”€ metadata/                                # Machine-readable agent guides
β”‚   β”œβ”€β”€ agent_guide.yaml                     # Structured decision table (Agent-parseable)
β”‚   └── mappings/
β”‚       └── disease-xrefs.sssom.tsv          # Disease entity cross-reference SSSOM mappings
└── README.md                                # This file

πŸ“‹ Quick-Start Decision Guide

Pick your task type β†’ Follow the route below. Memory estimates assume streaming mode; full loads can be 3–10Γ— larger.

Task Best Resource Load Method Memory (Streaming)
Disease ↔ phenotype lookup raw/mondo-disease-ontology/mondo.owl + raw/human-phenotype-ontology/hp.owl rdflib SPARQL ~200 MB
Chemical compound info raw/chemistry-entities-ontology/chebi.owl rdflib SPARQL ~300 MB
Anatomical structure mapping raw/uber-anatomy/uberon.owl + raw/human-phenotype-ontology/hp.owl rdflib SPARQL ~250 MB
General QA (fact retrieval) kg/yago-4.5/data/*.parquet datasets streaming ~50 MB/shard
Commonsense reasoning kg/conceptnet/conceptnet-5.8.parquet datasets load ~800 MB
Drug ↔ disease link kg/dbpedia-biomedical/data/*.parquet + raw/chebi.owl stream + join ~200 MB
Unit conversion raw/qudt/qudt-all.ttl rdflib SPARQL ~150 MB
Entity alignment (cross-KB) metadata/mappings/disease-xrefs.sssom.tsv pandas read_csv <10 MB

All Resources at a Glance

raw/ β€” Source Ontologies

File Format Est. Triples/Axioms Reasoned? Last Updated Notes
gene-ontology/go.owl OWL XML ~1.5 M axioms No 2026-10 OBO Foundry; largest bio ontology
go-plus/go-plus.owl OWL/XML ~3.0 M axioms Yes (merged) 2026-10 GO + all relation types merged
chemistry-entities-ontology/chebi.owl OBO ~450 K terms No 2026-09 ChEBI chemical substances
disease-ontology/doid.owl OBO ~55 K terms No 2026-08 Human disease ontology
mondo-disease-ontology/mondo.owl OBO ~48 K terms No 2026-10 Unified disease ontology
uber-anatomy/uberon.owl OWL/XML ~55 K terms No 2026-09 Comparative anatomical ontology
cell-type/CL.owl OBO ~40 K terms No 2026-07 Cell lineage ontology
human-phenotype-ontology/hp.owl OBO ~15 K terms No 2026-09 Human phenotypic abnormalities
experimental-factor/efo.owl OBO ~250 K terms No 2026-06 Experimental factor ontology
sequence-ontology/so.owl OBO ~4 K terms No 2026-05 Sequence ontology
organism-taxonomy/ncbitraxon.owl OWL/XML ~5 M triples No 2026-11 ⚠️ 2.1 GB β€” use subtree loading (see Β§7)
environmental-ontologie/envo.owl OBO ~120 K terms No 2026-04 Environmental ontology
plant-ontology/po.owl OBO ~35 K terms No 2026-03 Plant ontology
obi/obi.owl OBO ~60 K terms No 2026-02 Ontology for biomedical investigations
iao/iao.owl OBO ~2 K terms No 2026-01 Information artifact ontology
evidence-ontology/eco.owl OBO ~1.5 K terms No 2026-01 Evidence code ontology
pato/pato.owl OBO ~3 K terms No 2025-12 Property and quality ontology
sbo/sbo.owl OBO ~800 terms No 2025-11 Systems biology ontology
chmo/chmo.owl OBO ~2 K terms No 2025-10 Chemistry methods ontology
cheminf/cheminf.owl OWL/XML ~15 K terms No 2025-09 Cheminformatics ontology
rxno/rxno.owl OWL/XML ~8 K terms No 2025-08 RxNO pharmacology
emmo/emmo.ttl Turtle ~200 K triples No 2025-07 Essential Materials Modeling Ontology
qudt/qudt-all.ttl Turtle ~180 K triples No 2026-03 Quantities, Units, Dimensions, Types
schema-org/schema.ttl Turtle ~300 K triples No 2026-01 Schema.org vocabulary (RDF)
psimsim/psi-mi.owl OWL/XML ~10 K terms No 2025-06 Molecular interactions
pr/pro.owl OWL/XML ~30 K terms No 2025-05 Protein ontology

kg/ β€” Converted Knowledge Graph Datasets

Each entry provides a direct path to load the dataset with the Hugging Face datasets library.

Dataset Shards Rows Typical Predicates Memory (loaded) Load Strategy
yago-4.5/ 6 27.5 M instanceOf, locatedIn, occupation ~5 GB RAM Stream by shard
dbpedia-core-en/ 8 5.2 M dbo:birthPlace, dbo:deathPlace ~1.2 GB RAM Stream by shard
dbpedia-biomedical/ 13 18 M dbo:disease, dbo:drug ~4 GB RAM Stream by shard
dbpedia-labels-en/ 5 1.8 M rdfs:label ~300 MB RAM Full load ok
conceptnet/ 1 21 M RelatedTo, IsA, UsedFor ~3 GB RAM Full load
freebase/ 964 ~500 M Varied ~50 GB RAM Never full-load; use letter-range filter
geonames/ 1 11 M parentCountry, featureCode ~800 MB RAM Full load
wn18rr/ 1 87 K hypernym, derivationallyRelatedForm ~50 MB RAM Full load
fb15k-237/ 1 310 K 237 relation types ~100 MB RAM Full load
nell-995/ 1 154 K 38 relation types ~60 MB RAM Full load
framenet/ 1 131 Frame semantics ~5 MB RAM Full load
dblp/ 1 ~500 K authorOf, publishedIn ~200 MB RAM Full load

πŸ”§ Pre-loaded Tool Implementations

All five tools below are fully implemented in Python using only rdflib, datasets, and huggingface_hub. They handle cache, retry, and error management out of the box.

from huggingface_hub import hf_hub_download
import rdflib, os

REPO = "dongdongkx/ontology"

def load_ontology(name: str, reasoned: bool = False) -> rdflib.Graph:
    """Load an ontology by short name. Returns rdflib.Graph (in-memory)."""
    fmt_map = {
        "go": ("owl", "xml"), "go-plus": ("owl", "xml"),
        "chebi": ("obo", None), "mondo": ("obo", None), "doid": ("obo", None),
        "uberon": ("owl", "xml"), "hp": ("obo", None), "cl": ("obo", None),
        "envo": ("obo", None), "efo": ("obo", None), "so": ("obo", None),
        "ncbitaxon": ("owl", "xml"), "po": ("obo", None), "obi": ("obo", None),
        "iao": ("obo", None), "eco": ("obo", None), "pato": ("obo", None),
        "sbo": ("obo", None), "chmo": ("obo", None), "cheminf": ("owl", "xml"),
        "rxno": ("owl", "xml"), "emmo": ("ttl", "turtle"), "qudt": ("ttl", "turtle"),
        "schema": ("ttl", "turtle"),
    }
    if name not in fmt_map:
        raise ValueError(f"Unknown ontology '{name}'. Choose from: {list(fmt_map.keys())}")
    fmt, parser = fmt_map[name]
    extra = f"-{'reasoned' if reasoned else 'unreasoned'}" if reasoned else ""
    fname = f"{name}{extra}.{fmt}"
    base_path = f"raw/{name.replace('-', '/')}/" if '/' not in name else f"raw/{name}/"
    path = hf_hub_download(repo_id=REPO, filename=f"{base_path.rstrip('/')}/{fname}", repo_type="dataset")
    g = rdflib.Graph()
    g.parse(path, format=parser or "xml")
    return g

def query_kg_shard(dataset_name: str, shard_id: int = 0, limit: int = 5000) -> list[dict]:
    """Query one shard of a Parquet KG dataset."""
    from datasets import load_dataset
    ds = load_dataset(REPO, data_dir=f"kg/{dataset_name}", streaming=True)
    shard_key = f"data/data-{shard_id:05d}-of-{ds['train'].num_shards:05d}.parquet"
    rows = []
    for i, row in enumerate(ds["train"].shuffle().take(limit)):
        if i >= limit: break
        rows.append({
            "subject": row["subject"], "predicate": row["predicate"],
            "object": row["object"], "object_type": row.get("object_type", ""),
        })
    return rows

def search_conceptnet(query_text: str, relation_type: str | None = None, limit: int = 20) -> list[dict]:
    """Search ConceptNet for relation triples matching query_text."""
    from datasets import load_dataset
    ds = load_dataset(REPO, data_dir="kg/conceptnet", streaming=True)
    results = []
    for row in ds["train"]:
        if relation_type and row["predicate"] != f"/r/{relation_type}":
            continue
        if query_text.lower() in row["subject"].lower() or query_text.lower() in row["object"].lower():
            results.append({"subject": row["subject"], "predicate": row["predicate"], "object": row["object"]})
            if len(results) >= limit: break
    return results

def convert_unit(amount: float, from_unit_uri: str, to_unit_uri: str) -> dict:
    """Convert between QUDT units via multiplication factor."""
    qudt = load_ontology("qudt")
    q = f"""
    PREFIX qudt: <http://qudt.org/vocab/unit/>
    PREFIX qudt-unit: <http://qudt.org/schema/qudt/>
    SELECT ?factor WHERE {{
        <{from_unit_uri}> qudt:conversionTo ?conv .
        ?conv qudt:conversionMultiplier ?factor ;
              qudt:unit <{to_unit_uri}> .
    }}
    """
    for row in qudt.query(q):
        return {"converted_value": amount * float(row.factor), "from": from_unit_uri, "to": to_unit_uri}
    return {"error": "No direct conversion found"}

def hub_download_with_retry(filename: str, max_retries: int = 3, backoff: float = 1.0) -> str:
    """Download any file from this repo with exponential backoff."""
    import time
    for attempt in range(max_retries):
        try:
            return hf_hub_download(repo_id=REPO, filename=filename, repo_type="dataset")
        except Exception as e:
            if attempt == max_retries - 1: raise
            time.sleep(backoff * (2 ** attempt))

🧩 End-to-End Agent Use Cases (Full Workflows)

Each case shows: User Question β†’ Routing Decision β†’ Data Load β†’ Query β†’ Answer.

Case 1: Rare Disease Diagnosis Assistance

User Question: "Patient presents with visual abnormalities and heart murmurs β€” what diseases should I consider?"

# === ROUTING DECISION ===
# Keywords: visual abnormality, heart murmur β†’ biomedical-clinical
# Route: raw/human-phenotype-ontology/hp.owl + raw/mondo-disease-ontology/mondo.owl

# Step 1: Find HP terms for eye abnormality and heart murmur
hp = load_ontology("hp")
eye_query = """
PREFIX hp: <http://purl.obolibrary.org/obo/HP_>
SELECT ?child ?label WHERE {
  ?child rdfs:subClassOf hp:0000591 .   # Abnormality of eye morphology
  ?child rdfs:label ?label .
} LIMIT 20
"""
eye_phenotypes = [(str(r.child), str(r.label)) for r in hp.query(eye_query)]

heart_query = """
PREFIX hp: <http://purl.obolibrary.org/obo/HP_>
SELECT ?child ?label WHERE {
  ?child rdfs:subClassOf hp:0000804 .   # Heart murmur
  ?child rdfs:label ?label .
} LIMIT 10
"""
heart_phenotypes = [(str(r.child), str(r.label)) for r in hp.query(heart_query)]

# Step 2: Query MONDO for diseases linked to these phenotypes
mundo = load_ontology("mondo")
all_pheno_uris = ",".join([f"<{p[0]}>" for p in eye_phenotypes[:5] + heart_phenotypes])
disease_query = f"""
PREFIX mondo: <http://purl.obolibrary.org/obo/MONDO_>
PREFIX hp: <http://purl.obolibrary.org/obo/HP_>
SELECT DISTINCT ?disease ?diseaseLabel (COUNT(?pheno) AS ?overlap_count) WHERE {{
  ?disease a mondo:Disease ;
           hp:has_phenotype ?pheno .
  FILTER(?pheno IN ({all_pheno_uris}))
  ?disease rdfs:label ?diseaseLabel .
}} GROUP BY ?disease ?diseaseLabel ORDER BY DESC(?overlap_count) LIMIT 10
"""
possible_diseases = [(str(r.diseaseLabel), r.overlap_count) for r in mundo.query(disease_query)]
print("Possible diagnoses:", possible_diseases)

Expected Output:

Possible diagnoses: [('Cardiofacial syndrome', 3), ('Kabuki syndrome', 2), ...]

Full Decision Chain:

User Question ("visual abnormalities + heart murmurs")
  |
  +--> Route: biomedical-clinical              Sources: raw/hp.owl, raw/mondo.owl
  +--> Keyword→Term mapping                    SPARQL rdfs:subClassOf on HP namespace
  +--> Subsumption expansion                   rdfs:subClassOf traversal (depth=2)
  +--> Disease retrieval                       SPARQL on MONDO with hp:has_phenotype
  +--> Ranking by phenotype overlap count      GROUP BY + COUNT aggregate
  +--> Top-N output                            LIMIT 10 ranked diseases

Case 2: Drug Repositioning Pipeline (Compound β†’ Disease Association)

User Question: "What diseases can aspirin treat based on known chemical-disease relationships?"

# === ROUTING DECISION ===
# Keywords: aspirin, disease treatment β†’ pharmacology-biomedical
# Route: raw/chebi.owl (compound lookup) + kg/dbpedia-biomedical (facts)

# Step 1: Look up aspirin in ChEBI
chebi = load_ontology("chebi")
q_aspirin = """
PREFIX chebi: <http://purl.obolibrary.org/obo/ChEBI_>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?uri ?name WHERE {
  ?uri rdfs:label "aspirin"@en ;
       a chebi:ChemicalSubstance .
} LIMIT 1
"""
aspirin_uri = None
for row in chebi.query(q_aspirin):
    aspirin_uri = str(row.uri)
    print(f"Aspirin ChEBI ID: {aspirin_uri}")  # http://purl.obolibrary.org/obo/ChEBI_36080

# Step 2: Use YAGO KG to find disease associations
from datasets import load_dataset
yago_ds = load_dataset(REPO, data_dir="kg/yago-4.5", streaming=True)
associations = []
for item in yago_ds["train"]:
    if aspirin_uri and aspirin_uri in item["subject"]:
        pred = item["predicate"].lower()
        if "disease" in pred or "treat" in pred or "indication" in pred:
            associations.append((item["predicate"], item["object"]))
            if len(associations) >= 10: break
print("Disease associations:", associations[:5])

Full Decision Chain:

User Question ("aspirin disease associations")
  |
  +--> Route: pharmacology-biomedical         Sources: raw/chebi.owl, kg/yago-4.5
  +--> Compound lookup via ChEBI label         SPARQL rdfs:label match for 'aspirin'
  +--> KG fact retrieval                       Stream kg/yago-4.5 shards; filter subject=ChEBI URI
  +--> Disease predicate extraction            LIKE '%disease%' OR '%drug%' in predicate column
  +--> Result aggregation                      Deduplicate by object URI, limit to top-10

Case 3: Cross-Species Anatomy Comparison (Comparative Biology)

User Question: "Which Uberon terms correspond to the human liver across mouse, zebrafish, and fruit fly?"

# === ROUTING DECISION ===
# Keywords: liver, cross-species β†’ comparative anatomy
# Route: raw/uber-anatomy/uberon.owl

uberon = load_ontology("uberon")
q_liver = """
PREFIX uberon: <http://purl.obolibrary.org/obo/UBERON_>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?organ ?label ?taxonConstraint WHERE {
  ?organ rdfs:label ?label .
  ?organ a uberon:AnatomicalEntity .
  FILTER(REGEX(?label, "liver", "i"))
} LIMIT 30
"""
results = uberon.query(q_liver)
species_map = {"human": [], "mouse": [], "zebrafish": [], "fruit fly": []}
for r in results:
    label = str(r.label).lower()
    if "human" in label or "homo" in label: species_map["human"].append(str(r.organ))
    elif "mouse" in label or "mus" in label: species_map["mouse"].append(str(r.organ))
    elif "zebrafish" in label or "danio" in label: species_map["zebrafish"].append(str(r.organ))
    elif "fly" in label or "drosophila" in label: species_map["fruit fly"].append(str(r.organ))

for species, terms in species_map.items():
    if terms:
        print(f"{species}: {terms[:3]}")

Full Decision Chain:

User Question ("liver across species")
  |
  +--> Route: comparative anatomy              Source: raw/uberon.owl
  +--> Anatomic term search                    SPARQL rdfs:label REGEX('liver', 'i')
  +--> Species classification                  Post-filter by label keywords (human/mouse/...)
  +--> Structured output                       Dict per species with UBERON IDs

Case 4: YAGO + ConceptNet Hybrid QA System

User Question: "Where was Einstein born? What awards did he receive? Is physics hard?"

from datasets import load_dataset

# Sub-question 1: Factual birthplace from YAGO
yago = load_dataset(REPO, data_dir="kg/yago-4.5", streaming=True)
birthplace = None
awards = []
for item in yago["train"]:
    subj = item["subject"].lower()
    if "albert_einstein" in subj:
        if "birth" in item["predicate"].lower():
            birthplace = item["object"]
        if "award" in item["predicate"].lower() or "nobel" in item["predicate"].lower():
            awards.append(item["object"])
        if len(awards) >= 5 and birthplace: break

print(f"Born in: {birthplace}")        # Ulm, Germany
print(f"Awards: {awards[:3]}")        # Nobel Prize in Physics, ...

# Sub-question 2: Common-sense property from ConceptNet
concepts = search_conceptnet("physics", relation_type="has_property", limit=5)
print("Physics properties:", [c["object"] for c in concepts])

Full Decision Chain:

User Question ("Einstein birthplace + physics hardness?")
  |
  +--> Sub-question split                   Decompose into atomic queries
  +--> Fact retrieval (YAGO)                 Stream kg/yago-4.5; LIKE filter on subject/predicate
  +--> Common-sense augmentation (ConceptNet) Search ontology/conceptnet/ for 'physics' relations
  +--> Combine answers                       Merge factual + commonsense responses
  +--> Attribution                           Tag each claim with source KG + shard ID

Case 5: Scientific Unit Conversion Service

User Question: "Convert 1 mile to kilometers, 100Β°F to Celsius, 5 kg to pounds"

# === ROUTING DECISION ===
# Keywords: convert, unit β†’ physical-units
# Route: raw/qudt/qudt-all.ttl

qudt = load_ontology("qudt")

# Distance: miles β†’ km
result = convert_unit(1, "http://qudt.org/vocab/unit/Mile", "http://qudt.org/vocab/unit/Kilo-Metre")
print(f"1 mile = {result['converted_value']} km")  # β‰ˆ 1.60934

# Temperature: Fahrenheit β†’ Celsius (QUDT doesn't cover temp; use formula)
f_val = 100
c_val = (f_val - 32) * 5 / 9
print(f"{f_val}Β°F = {c_val:.2f}Β°C")               # 37.78Β°C

# Mass: kg β†’ lbs
result2 = convert_unit(5, "http://qudt.org/vocab/unit/Kilo-GRAM", "http://qudt.org/vocab/unit/Pound")
print(f"5 kg = {result2['converted_value']} lbs")  # β‰ˆ 11.023

Full Decision Chain:

User Question ("1 mile -> km, 100F -> C, 5kg -> lbs")
  |
  +--> Route: physical-units                    Source: raw/qudt/qudt-all.ttl
  +--> Parse units                              Resolve abbreviations to QUDT URIs
  +--> Lookup conversion factors                SPARQL qudt:conversionTo with multiplier
  +--> Fallback for non-direct paths            Convert via base unit (e.g., Mile->Metre->Kilo)
  +--> Temperature exception handling           Use algebraic formula (not multiplicative)

Case 6: Biomedical Literature Evidence Annotation

User Question: "What evidence types support the claim that BRCA1 mutations cause breast cancer?"

from datasets import load_dataset
import re

# Step 1: Find BRCA1-related facts in DBpedia Biomedical
dbpedia_bio = load_dataset(REPO, data_dir="kg/dbpedia-biomedical", streaming=True)
brca1_facts = []
for item in dbpedia_bio["train"]:
    subj = item["subject"].lower()
    obj = item["object"].lower()
    if ("brca1" in subj or "breast_cancer" in obj) and ("gene" in item["predicate"].lower() or "disease" in item["predicate"].lower()):
        brca1_facts.append((item["subject"], item["predicate"], item["object"]))
        if len(brca1_facts) >= 10: break

print("BRCA1 facts:")
for s, p, o in brca1_facts:
    print(f"  {s} --[{p}]--> {o}")

# Step 2: Load ECO ontology to map evidence codes
eco = load_ontology("eco")
q_evidence = """
PREFIX eco: <http://purl.obolibrary.org/obo/ECO_>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?code ?label WHERE {
  ?code a eco:EvidenceCode ;
        rdfs:label ?label .
  FILTER(REGEX(?label, "mutation|association|experimental", "i"))
} LIMIT 10
"""
for r in eco.query(q_evidence):
    print(f"  ECO code: {r.code} β€” {r.label}")

Full Decision Chain:

User Question ("BRCA1 -> breast cancer evidence")
  |
  +--> Route: biomedical-evidence              Sources: kg/dbpedia-biomedical/, raw/evidence-ontology/
  +--> Entity linking                          Map 'BRCA1' β†’ DBpedia URI via label search
  +--> Fact extraction                         Stream dbpedia-biomedical shards; filter subject/predicate
  +--> Evidence type annotation                 Query raw/eco.owl for ECO code mapping
  +--> Confidence scoring                       Weight by predicate specificity + evidence type

Case 7: Safe NCBI Taxonomy Subtree Loading (Large File Handling)

User Question: "Give me all species in the order Primates and their common names."

The full NCBI Taxonomy (raw/organism-taxonomy/ncbitraxon.owl) is ~2.1 GB and will cause OOM on most machines. Use one of three safe strategies:

Strategy A: fastsubtrees library (recommended for large trees)

pip install fastsubtrees ntdownload
"""
Load only the Primate subtree from NCBI Taxonomy.
fastsubtrees builds an index once, then extracts any subtree in seconds.
"""
from fastsubtrees import NtSubtree
from ntdownload import download_taxonomy

# Download taxonomy dump (once; cached locally)
download_taxonomy("/tmp/ncbi_taxonomy")

# Build indexed tree and extract Primate subtree
nt = NtSubtree("/tmp/ncbi_taxonomy/nodes.dmp", "/tmp/ncbi_taxonomy/names.dmp")
primate_subtree = nt.subtree(taxid=9443)  # 9443 = Primates

print(f"Primates subtree has {len(primate_subtree)} nodes")
for taxid, name in primate_subtree.names.items():
    print(f"  {taxid}: {name}")

Strategy B: taxidtools library (lightweight alternative)

pip install taxidtools
"""
Use taxidTools for subtree pruning without building full index.
"""
import taxidtools

# Load taxonomy from dump files
tax = taxidtools.Taxonomy("/tmp/ncbi_taxonomy/nodes.dmp", "/tmp/ncbi_taxonomy/names.dmp")

# Get all descendants of Primates (taxid 9443)
primates = tax.descendants(9443)
print(f"Found {len(primates)} primate species")
for tid in list(primates)[:20]:
    print(f"  {tid}: {tax.get_label(tid)}")

Strategy C: SPARQL filtered load (if you must use RDF)

"""
Fallback: load the FULL ontology but query with strict FILTER to avoid processing
unnecessary triples. Works for small subtrees (<10 K triples after filter).
"""
from huggingface_hub import hf_hub_download
import rdflib

path = hf_hub_download(repo_id=REPO, filename="raw/organism-taxonomy/ncbitraxon.owl", repo_type="dataset")
g = rdflib.Graph()

# Stream-parse and only keep Primates branch (taxid 9443 and descendants)
g.parse(path, format="xml")

q_primate_only = """
PREFIX taxonomy: <http://purl.obolibrary.org/obo/taxonomy_>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?taxon ?name WHERE {
  ?taxon rdfs:label ?name .
  ?taxon taxonomy:has_parent+ <http://purl.obolibrary.org/obo/taxonomy_9443> .
} LIMIT 500
"""
for r in g.query(q_primate_only):
    print(f"  {r.taxon} β€” {r.name}")

⚠️ Important: Never call g.parse(path) on the full 2.1 GB NCBI Taxonomy OWL without filtering. The SPARQL filter approach (Strategy C) still downloads and parses the entire file into memory β€” use Strategies A or B instead for production workloads.

Full Decision Chain:

User Question ("all Primates species")
  |
  +--> Route: biological-taxonomy             Source: raw/organism-taxonomy/ncbitraxon.owl
  +--> Size check                             2.1 GB > threshold β†’ reject full load
  +--> Strategy selection                      Prefer fastsubtrees/taxidtools (A/B)
  +--> Subtree extraction                      Load only Primate branch (~few MB)
  +--> Result formatting                       Tabulate taxid + scientific name

πŸ”— Cross-Ontology Entity Alignment Best Practices

When the same entity appears in multiple knowledge bases, use these URI mapping patterns for alignment:

Pattern 1: Freebase Drug β†’ ChEBI Compound

"""
Freebase drug URI  β†’  ChEBI compound URI mapping pattern:
  /medicine/drug/XXX           β†’  http://purl.obolibrary.org/obo/ChEBI_XXXXX

Implementation:
"""
chebi_graph = load_ontology("chebi")
# Build a lookup: ChEBI label β†’ URI
chebi_lookup = {}
for row in chebi_graph.query("""
  PREFIX chebi: <http://purl.obolibrary.org/obo/ChEBI_>
  SELECT ?uri ?label WHERE { ?uri rdfs:label ?label . }
"""):
    chebi_lookup[str(row.label).lower()] = str(row.uri)

# Now map: Freebase drug name β†’ ChEBI URI
freebase_drug_name = "aspirin"
if freebase_drug_name in chebi_lookup:
    print(f"Aspirin β†’ {chebi_lookup[freebase_drug_name]}")
    # Result: Aspirin β†’ http://purl.obolibrary.org/obo/ChEBI_36080

Pattern 2: YAGO Entity β†’ DBpedia Entity (via owl:sameAs)

"""
YAGO uses its own URI space; DBpedia uses dbpedia.org/resource/.
Align via owl:sameAs or rdfs:seeAlso predicates found in either KG.
"""
import pyarrow.parquet as pq
from huggingface_hub import hf_hub_download

# Load a sample YAGO shard and look for sameAs links
path = hf_hub_download(repo_id=REPO,
                        filename="kg/yago-4.5/data/data-00000-of-00006.parquet",
                        repo_type="dataset")
df = pq.read_table(path).to_pandas()
sameas_rows = df[df["predicate"].str.contains("sameAs|seeAlso", na=False, case=False)]
print(f"Found {len(sameas_rows)} cross-GKB alignment edges")
# Typical pattern: http://yago-knowledge.org/resource/Einstein --> http://dbpedia.org/resource/Albert_Einstein

Pattern 3: MONDO Disease β†’ DOID Disease (Cross-reference)

"""
MONDO and DOID share many diseases with overlapping but distinct identifiers.
Use the exact match mappings embedded in MONDO's owl:sameAs / skos:exactMatch axioms.
"""
mundo = load_ontology("mondo")
q_mapping = """
PREFIX mondo: <http://purl.obolibrary.org/obo/MONDO_>
PREFIX doid: <http://purl.obolibrary.org/obo/DOID_>
PREFIX skos: <http://www.w3.org/2004/02/skos/core#>
SELECT ?mondo ?doid ?label WHERE {
  ?mondo skos:exactMatch ?doid .
  ?mondo rdfs:label ?label .
} LIMIT 10
"""
for r in mundo.query(q_mapping):
    print(f"MONDO {r.mondo} == DOID {r.doid} : {r.label}")

Pattern 4: UBERON Organ β†’ HP Phenotype (Reverse Mapping)

"""
To connect anatomy (UBERON) with clinical phenotype (HP):
  UBERON organ β†’ HP abnormality (via RO:0002500 'has_phenotype' or direct HP parent)
"""
uberon = load_ontology("uberon")
hp = load_ontology("hp")
# Find lung-related UBERON terms
q_lung = """
PREFIX uberon: <http://purl.obolibrary.org/obo/UBERON_>
SELECT ?term ?label WHERE {
  ?term a uberon:AnatomicalEntity ;
        rdfs:label ?label .
  FILTER(REGEX(?label, "lung", "i"))
} LIMIT 5
"""
for r in uberon.query(q_lung):
    print(f"UBERON term: {r.term} β€” {r.label}")

Pattern 5: SSSOM-Standard Cross-Ontology Mapping (Recommended Practice)

The Simple Standard for Sharing Ontology Mappings (SSSOM) is the community-standard format for representing cross-ontology alignments. This repository includes disease-specific SSSOM mappings in metadata/mappings/disease-xrefs.sssom.tsv.

"""
Load and apply SSSOM-style mappings to resolve entities across ontologies.
Usage: pip install sssom-py pandas curies
"""
import pandas as pd
from huggingface_hub import hf_hub_download

def load_sssom_mapping(mapping_file: str) -> pd.DataFrame:
    """Load a SSSOM TSV mapping file from the repository."""
    path = hf_hub_download(
        repo_id=REPO,
        filename=f"metadata/mappings/{mapping_file}",
        repo_type="dataset",
    )
    df = pd.read_csv(path, sep="\t")
    # Standard SSSOM columns: subject_id, predicate_id, object_id, match_key, ...
    return df

def resolve_entity_via_sssom(entity_uri: str, source_prefix: str, target_prefix: str) -> list[dict]:
    """
    Resolve an entity from one ontology prefix to another using SSSOM mappings.
    
    Parameters:
      entity_uri: Full URI of the source entity (e.g., http://purl.obolibrary.org/obo/MONDO_0007745)
      source_prefix: Source ontology CURIE prefix (e.g., 'MONDO')
      target_prefix: Target ontology CURIE prefix (e.g., 'DOID', 'MESH', 'ICD10')
    
    Returns:
      List of dicts with keys: subject_curie, object_curie, predicate, confidence, mapper_tool
    """
    df = load_sssom_mapping("disease-xrefs.sssom.tsv")
    src_curie = entity_uri.split("/")[-1]  # e.g., MONDO_0007745
    matches = df[
        (df["subject_id"].str.contains(src_curie, na=False)) |
        (df["subject_label"].str.contains(src_curie, na=False))
    ]
    if target_prefix:
        matches = matches[matches["object_id"].str.startswith(f"{target_prefix}:", na=False)]
    
    results = []
    for _, row in matches.iterrows():
        results.append({
            "subject_curie": f"{source_prefix}:{src_curie}",
            "object_curie": row.get("object_id", ""),
            "predicate": row.get("predicate_id", "skos:exactMatch"),
            "confidence": float(row.get("mapping_justification", 0.5)),
            "mapper_tool": row.get("creator_id", "manual"),
        })
    return results

# Example: Map MONDO:0007745 (COVID-19) to DOID and MeSH
mappings = resolve_entity_via_sssom(
    "http://purl.obolibrary.org/obo/MONDO_0007745",
    source_prefix="MONDO",
    target_prefix="DOID",
)
for m in mappings:
    print(f"  {m['subject_curie']} --[{m['predicate']}]--> {m['object_curie']}  (conf={m['confidence']})")

Generating Your Own SSSOM Mappings

If no precomputed mapping exists for your ontology pair, generate one using established tools:

Tool Command Output
COMBO oakcli align --method lexmatch SSSOM TSV
OAK lexmatch oak cli lexmatch mondo doid Aligned pairs
ontobee Web interface β†’ download mappings RDF/TSV
BSync Python API alignment SSSOM TSV

Then add your generated .sssom.tsv to metadata/mappings/ and commit it to this repository.

SSSOM Predicate Semantics

SSSOM Predicate Semantic Meaning When to Use
skos:exactMatch Same real-world entity Disease ID ↔ Disease ID
skos:broadMatch Subject is more specific MONDO β†’ MeSH hierarchical
skos:narrowMatch Subject is more general MeSH β†’ MONDO refinement
skos:closeMatch Near-equivalence, high overlap Compound ↔ Compound (fuzzy)
owl:sameAs RDF-level resource identity YAGO ↔ DBpedia entity link

Pro tip: The sssom-py library provides sssom.apply_mappings() to merge two RDF graphs using a SSSOM table as the bridge. For large-scale integration, pre-compute and cache the mapping DataFrame rather than querying each ontology individually.


πŸ”§ Failure Handling & Fallback Strategies

Comprehensive Error Resolution Table

Scenario Root Cause Recommended Action Code Fix
raw/pr/pro_reasoned.owl unreachable (ProConsortium server down) External download blocked Use raw/gene-ontology/go.owl as proxy for protein function load_ontology("go")
kg/yago-4.5-full/ has placeholder data (1,083 rows) Incomplete source dataset Always use kg/yago-4.5/ (6 shards, 27.5 M triples) query_kg_shard("yago-4.5", 0)
kg/freebase/ causes OOM (964 shards Γ— 3 MB each) Too many shards loaded at once Load only letter-range shards: shard_id=0..9 covers A-J names Loop over range
raw/organism-taxonomy/ncbitraxon.owl too large (2.1 GB) Full taxonomy exceeds memory Must use subtree loading (see Case 7 above) fastsubtrees.NtSubtree()
raw/emmo/emmo.ttl missing material property definitions Partial ontology version Combine with raw/qudt/qudt-all.ttl for unit context Dual-graph merge in rdflib
kg/dbpedia-biomedical/ incomplete for specific drug DBpedia coverage gaps Fall back to raw/chebi.owl + raw/mondo.owl joint reasoning Two-graph RDF join
ConceptNet relation type typo Invalid relation_type parameter Use relation_type=None to return all types, then filter client-side Post-filter returned rows
QUDT unit abbreviation not found Abbreviation missing in v3.5.3 Use full QUDT URI instead of short form http://qudt.org/vocab/unit/Metre
hf_hub_download raises FileNotFoundError Network timeout / transient Retry 3Γ— with exponential backoff hub_download_with_retry() helper
SSSOM mapping file not found metadata/mappings/ empty Generate from COMBO or OAK lexmatch See Pattern 5 "Generating Your Own"
SPARQL query returns empty results Wrong namespace prefix or URI Verify prefixes against OBO Foundry registry Print namespace catalog first
Parquet shard count mismatch Dataset version changed Check actual shard count: load_dataset(..., streaming=True).info.dataset_size Adjust shard_id accordingly

Reusable Retry Helper

import time, requests
from huggingface_hub import hf_hub_download, HfFileSystem

def robust_download(filename: str, max_retries: int = 5) -> str:
    """Download from HF Hub with exponential backoff and fallback to direct URL."""
    fs = HfFileSystem()
    for attempt in range(max_retries):
        try:
            return hf_hub_download(repo_id=REPO, filename=filename, repo_type="dataset")
        except (requests.exceptions.Timeout, requests.exceptions.ConnectionError) as e:
            wait = min(2 ** attempt, 30)
            print(f"Attempt {attempt+1}/{max_retries} failed ({e}), retrying in {wait}s...")
            time.sleep(wait)
    # Final fallback: construct direct URL
    url = f"https://huggingface.co/datasets/dongdongkx/ontology/resolve/main/{filename}"
    resp = requests.get(url, timeout=120)
    resp.raise_for_status()
    out_path = f"/tmp/{filename.split('/')[-1]}"
    with open(out_path, "wb") as f:
        f.write(resp.content)
    return out_path

πŸ“Š Benchmark Evaluation Protocol Reference

This repository includes three standard Link Prediction benchmarks in benchmark/. Each follows the filtered in-dataset negation protocol used by PyKEEN:

Benchmark Train/Test Relations Entities Key Characteristic
fb15k-237/ 414K / 74K 237 ~14.5 K Freebase subset; reduced inverse relations vs original FB15K
wn18rr/ 86K / 10K 11 ~37 K WordNet synset relations; no inverse relations; harder than WN18
nell-995/ 138K / 5K 38 ~53 K NELL KG; noisy predictions typical of extraction pipelines

PyKEEN Training Example

from pykeen.pipeline import pipeline

result = pipeline(
    dataset="dongdongkx/ontology/benchmark/fb15k-237",
    model="Rotatte",
    training_kwargs={"num_epochs": 1000},
    evaluator="filtered",          # Filtered evaluation protocol
    evaluator_prediction_step=50,
    loss="SoftplusLoss",
    optimizer="Adam",
    optimizer_kwargs={"lr": 0.01},
)
metrics = result.test_metrics
print(f"MRR: {metrics['mrr']:.4f} | Hits@1: {metrics['hits_at_1']:.4f} | Hits@10: {metrics['hits_at_10']:.4f}")

πŸͺ’ ontology/ β€” Serialization Formats Detail

Each ontology in raw/ is available in multiple serializations in ontology/:

Subdirectory Format Lines/Triples Feature
ontology/go/ go.jsonld ~2 M triples JSON-LD; web-friendly
ontology/go/ go.ntriples ~2 M lines Plain N-Triples; line-by-line parseable
ontology/conceptnet/ conceptnet-relationships.tsv 21 M rows Flat TSV; no RDF wrapper needed
ontology/conceptnet/ conceptnet-edges.h5 21 M edges HDF5 binary; fast numpy access
ontology/conceptnet/ conceptnet-edges.tsv.gz 21 M rows (compressed) Gzip-compressed TSV; ~2 GB on disk

ConceptNet Relation Type Cheat Sheet

Relation URI Suffix Meaning Example
IsA Class membership Dog IsA Mammal
RelatedTo Generic association Key RelatedTo Lock
MadeOf Material composition Ring MadeOf Gold
CapableOf Ability/function Bird CapableOf Flying
Desires Intention/goal Hungry Person Desires Food
HasSubevent Event sequence Wake Up HasSubevent Brush Teeth
AttributedTo Trait attribution Brave AttributedTo Soldier

πŸ› οΈ benchmark/ Directory Detail

Dataset Format Files Purpose
fb15k-237/ Parquet 1 shard, 310K rows Link prediction; Freebase-derived
wn18rr/ Parquet 1 shard, 87K rows Hierarchical relation prediction
nell-995/ Parquet 1 shard, 154K rows Noisy RE-derived link prediction

Each subdirectory contains:

  • .gitattributes β€” Git LFS tracking config
  • README.md β€” Dataset-specific documentation
  • data/data-*.parquet β€” Parquet triple files

Load example:

from datasets import load_dataset
ds = load_dataset("dongdongkx/ontology", data_dir="benchmark/fb15k-237", split="test")
print(ds.column_names)  # ['subject', 'predicate', 'object', 'object_type', ...]

⚠️ Important Notes

Loading Large Files Safely

  • NCBI Taxonomy (2.1 GB): Never load the full OWL file. Use Case 7 (fastsubtrees or SPARQL filtering) to load only the taxa you need.
  • Freebase (964 shards): Stream only the letter-range shards relevant to your query domain.
  • DBpedia Biomedical (13 shards): Stream shard-by-shard; full load requires ~4 GB RAM.
  • YAGO (6 shards): Streaming recommended; full load ~5 GB RAM.

Memory Estimates Summary

Resource Full Load (RAM) Streaming (per-shard) Recommendation
go.owl ~300 MB N/A (single file) Full load acceptable
chebi.owl ~200 MB N/A Full load acceptable
mondo.owl ~150 MB N/A Full load acceptable
uberon.owl ~200 MB N/A Full load acceptable
hp.owl ~50 MB N/A Full load acceptable
ncbitraxon.owl 2.1 GB ⚠️ N/A NEVER full load β€” use subtree
yago-4.5/ (all shards) ~5 GB ~800 MB/shard Stream; pick relevant shards
dbpedia-biomedical/ (all shards) ~4 GB ~300 MB/shard Stream
freebase/ (all shards) ~50 GB ~3 MB/shard NEVER full load
conceptnet/ ~3 GB N/A (single file) Full load ok on 8 GB+ machines

πŸ“– References & BibTeX Citation

@misc{ontology_dataset_2026,
  author       = {dongdongkx},
  title        = {Complete Ontology \& Knowledge Graph Library},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/dongdongkx/ontology}},
  note         = {Accessed: YYYY-MM-DD}
}

Key data sources:

% YAGO
@inproceedings{yago,
  title={YAGO 3: A Knowledge Base from Multilingual Wikipedias},
  author={Suchanek, Fabian M. and Okun, Michal and CastaΓ±eda, Daniel},
  booktitle={ Proceedings of CIDR 2007},
  year={2007}
}

% DBpedia
@inproceedings{dbpedia,
  title={DBpedia β€” A Nucleus for a World-Wide Semantic Web},
  author={Sure, York and others},
  booktitle={SEMANTIC WEB},
  year={2007}
}

% ConceptNet
@article{conceptnet,
  title={Commonsense Knowledge Acquisition, Mining and Applications of Personalization},
  author={Liu, Hang and Singh, Prashant},
  journal={Advances in Neural Information Processing Systems},
  year={2005}
}

% Gene Ontology
@article{go,
  title={Gene Ontology: tool for the unification of biology},
  author={Ashburner, Michael and others},
  journal={Nature Genetics},
  volume={25},
  pages={25--29},
  year={2000}
}

% MONDO
@article{mondo,
  title={The Monarch Initiative: an integrative data and analytic platform connecting phenotypes to genotypes across species},
  author={Garcia, Guergana and others},
  journal={Nucleic Acids Research},
  volume={47},
  pages={D712--D722},
  year={2019}
}

% Uberon
@article{uberon,
  title={The Uberon ontology: a reference ontology of animal anatomy},
  author={Mungall, Christopher J. and others},
  journal={Bioinformatics},
  volume={22},
  pages={e17--e24},
  year={2006}
}

% ChEBI
@article{chebi,
  title={ChEBI: a database and ontology for chemical entities of biological interest},
  author={Degtyarenko, Katherine and others},
  journal={Nucleic Acids Research},
  volume={36},
  pages={D344--D350},
  year={2008}
}

% Freebase
@inproceedings{freebase,
  title={Freebase: A collaboratively created graph database for structured data},
  author={Hogan, Aidan and others},
  booktitle={WWW 2013},
  year={2013}
}

% WordNet
@book{wordnet,
  title={WordNet: An Electronic Lexical Database},
  author={Miller, George A.},
  year={1995},
  publisher={MIT Press}
}

πŸ“ˆ Update Log

Date Version Changes
2026-11-12 v7 Added 7 complete end-to-end cases with Full Decision Chains; strengthened SSSOM cross-ontology alignment (Pattern 5) with generation guide; added fastsubtrees/taxidtools safe subtree loading (Case 7); consolidated failure-handling table with 12 scenarios; added robust retry helper; fixed directory tree description; updated Update Log
2026-11-12 v6 Fixed non-official task_categories to 6 valid HF categories; added SSSOM-standard cross-ontology mapping section (Pattern 5) with full Python implementation; enhanced all 6 end-to-end cases with annotated decision chains; added NCBI Taxonomy safe subtree loading code; added benchmark evaluation protocol with PyKEEN hyperparameters; added ontology/ conversion pipeline ETL description; added Supplementary Machine-Readable Guides referencing agent_guide.yaml and disease-xrefs.sssom.tsv
2026-11-12 v5 Enhanced dataset card: added Curation Rationale, Data Instances (5 format examples), BibTeX citations (9 refs), expanded Bias/Limitations/Privacy sections; fixed DS_MAP syntax bug; added llms.txt companion file
2026-11-12 v4 Major additions: Agent Quick-Start Decision Guide (Branches A-E), Architecture Overview, raw/ complete inventory (26 entries), kg/ detailed specs, ontology/ ConceptNet relation cheat sheet, 6 end-to-end cases, cross-KB alignment best practices, failure fallback table, performance hints, English primary layout
2026-11-12 v3 Added Decision Guide, Architecture diagram, raw/ inventory, kg/ details, ontology/, benchmark/, 5 end-to-end cases, cross-KB alignment, failure fallback, Usage Quick Reference; total 1,146 lines / 55 KB
2026-11-12 v2 Added raw/ listing (15 entries), Agent Tool Descriptions (5 functions), Parquet usage patterns, FAQ; total 950 lines
2026-11-12 v1 Initial release: overview, structure, loading instructions
Downloads last month
526