66 · SemaTyP
Drug-Disease Association Knowledge Graph from literature mining + TTD
Category: Drug-centric | Type: KG | Subcategory: Drug-Disease Associations
Access: Local files (downloaded from GitHub)
What it provides
SemaTyP combines two data sources into a knowledge graph for drug discovery / repositioning:
- SemMedDB predications (
data/SemmedDB/predications.txt): subject–predicate–object triples with UMLS semantic types, extracted from PubMed abstracts (full version: ~39M triples).
- TTD curated associations (
data/TTD/): drug-disease and target-disease links from Therapeutic Target Database (2016), with ICD-9/ICD-10 codes.
- Processed associations (
data/processed/): pre-computed drug→disease, disease→drug, disease→target mappings.
Data schema
| File |
Format |
Columns |
data/SemmedDB/predications.txt |
TSV |
subject, object, context, predicate, subj_semtype, obj_semtype |
data/TTD/drug-disease_TTD2016.txt |
TSV |
TTDDRUGID, drug_name, indication, ICD9, ICD10 |
data/TTD/target-disease_TTD2016.txt |
TSV |
target_id, target_name, disease, ICD9, ICD10 |
data/processed/drug_disease |
TSV |
drug, disease, ... |
data/processed/disease_drug |
TSV |
disease, drug, ... |
data/processed/disease_targets |
TSV |
disease, target, ... |
Setup
Data must be downloaded locally first:
git clone https://github.com/ShengtianSang/SemaTyP.git
Then set the data path (default points to your HiPerGator location):
# Option 1: environment variable
export SEMATYP_DATA_DIR="/path/to/SemaTyP-main"
# Option 2: edit DATA_DIR in 66_SemaTyP.py directly
Note: The GitHub repo only contains a 100-line sample of predications.txt. The full 39M-triple file must be downloaded separately per the repo README.
Quick start
from 66_SemaTyP import query
# Single entity (drug, disease, target, or any biomedical concept)
results = query("aspirin")
# Multiple entities
results = query(["metformin", "diabetes", "CYP2D6"])
# Specific data source only
results = query("imatinib", fields="predications", pred_limit=10)
results = query("Schizophrenia", fields="ttd")
results = query("cancer", fields="processed")
query() interface
query(entities, fields="all", pred_limit=50) -> list[dict]
| Parameter |
Type |
Description |
entities |
str | list[str] |
Entity name(s), case-insensitive |
fields |
str |
"all" — everything; "predications" / "ttd" / "processed" |
pred_limit |
int |
Max predication triples per entity (default 50) |
Return structure (fields="all")
[
{
"query": "aspirin",
"predications": [
{"subject": "aspirin", "object": "pain", "predicate": "TREATS",
"context": "...", "subj_type": "phsu", "obj_type": "sosy"}
],
"predication_count": 42,
"ttd_drug_disease": [
{"ttdid": "DAP000XXX", "drug": "Aspirin", "disease": "Pain",
"icd9": "...", "icd10": "..."}
],
"ttd_target_disease": [],
"processed_drug_disease": [...],
"processed_disease_drug": [...],
"processed_disease_targets": [...]
}
]
Lower-level functions
| Function |
Input |
Output |
Description |
get_predications(entity, limit=50) |
entity name |
list[dict] |
SemMedDB KG triples |
get_ttd_drug_disease(entity) |
drug or disease |
list[dict] |
TTD drug-disease associations |
get_ttd_target_disease(entity) |
target or disease |
list[dict] |
TTD target-disease associations |
get_processed_drug_disease(entity) |
entity name |
list[dict] |
Processed drug→disease |
get_processed_disease_drug(entity) |
entity name |
list[dict] |
Processed disease→drug |
get_processed_disease_targets(entity) |
entity name |
list[dict] |
Processed disease→target |
Notes
- All lookups are case-insensitive (indexed by lowercased entity names).
- Data is lazy-loaded: first
query() call triggers a one-time index build (may take seconds for large predications file).
- UMLS semantic type codes in predications:
phsu = pharmaceutical substance, dsyn = disease/syndrome, gngm = gene/genome, sosy = sign/symptom, podg = patient/group, etc.
- The GitHub sample
predications.txt has only 100 lines; for full coverage, download the complete file per the repo README instructions.
1---2name: sematyp3description: 66 · SemaTyP4---5# 66 · SemaTyP67> Drug-Disease Association Knowledge Graph from literature mining + TTD 8> **Category:** Drug-centric | **Type:** KG | **Subcategory:** Drug-Disease Associations 9> **Access:** Local files (downloaded from GitHub)1011| Resource | URL |12|----------|-----|13| GitHub | https://github.com/ShengtianSang/SemaTyP |14| Paper | https://link.springer.com/article/10.1186/s12859-018-2167-5 |1516---1718## What it provides1920SemaTyP combines two data sources into a knowledge graph for drug discovery / repositioning:2122- **SemMedDB predications** (`data/SemmedDB/predications.txt`): subject–predicate–object triples with UMLS semantic types, extracted from PubMed abstracts (full version: ~39M triples).23- **TTD curated associations** (`data/TTD/`): drug-disease and target-disease links from Therapeutic Target Database (2016), with ICD-9/ICD-10 codes.24- **Processed associations** (`data/processed/`): pre-computed drug→disease, disease→drug, disease→target mappings.2526### Data schema2728| File | Format | Columns |29|------|--------|---------|30| `data/SemmedDB/predications.txt` | TSV | subject, object, context, predicate, subj_semtype, obj_semtype |31| `data/TTD/drug-disease_TTD2016.txt` | TSV | TTDDRUGID, drug_name, indication, ICD9, ICD10 |32| `data/TTD/target-disease_TTD2016.txt` | TSV | target_id, target_name, disease, ICD9, ICD10 |33| `data/processed/drug_disease` | TSV | drug, disease, ... |34| `data/processed/disease_drug` | TSV | disease, drug, ... |35| `data/processed/disease_targets` | TSV | disease, target, ... |3637---3839## Setup4041Data must be downloaded locally first:4243```bash44git clone https://github.com/ShengtianSang/SemaTyP.git45```4647Then set the data path (default points to your HiPerGator location):4849```python50# Option 1: environment variable51export SEMATYP_DATA_DIR="/path/to/SemaTyP-main"5253# Option 2: edit DATA_DIR in 66_SemaTyP.py directly54```5556**Note:** The GitHub repo only contains a 100-line sample of `predications.txt`. The full 39M-triple file must be downloaded separately per the repo README.5758---5960## Quick start6162```python63from 66_SemaTyP import query6465# Single entity (drug, disease, target, or any biomedical concept)66results = query("aspirin")6768# Multiple entities69results = query(["metformin", "diabetes", "CYP2D6"])7071# Specific data source only72results = query("imatinib", fields="predications", pred_limit=10)73results = query("Schizophrenia", fields="ttd")74results = query("cancer", fields="processed")75```7677---7879## `query()` interface8081```82query(entities, fields="all", pred_limit=50) -> list[dict]83```8485| Parameter | Type | Description |86|-----------|------|-------------|87| `entities` | `str \| list[str]` | Entity name(s), case-insensitive |88| `fields` | `str` | `"all"` — everything; `"predications"` / `"ttd"` / `"processed"` |89| `pred_limit` | `int` | Max predication triples per entity (default 50) |9091### Return structure (`fields="all"`)9293```json94[95 {96 "query": "aspirin",97 "predications": [98 {"subject": "aspirin", "object": "pain", "predicate": "TREATS",99 "context": "...", "subj_type": "phsu", "obj_type": "sosy"}100 ],101 "predication_count": 42,102 "ttd_drug_disease": [103 {"ttdid": "DAP000XXX", "drug": "Aspirin", "disease": "Pain",104 "icd9": "...", "icd10": "..."}105 ],106 "ttd_target_disease": [],107 "processed_drug_disease": [...],108 "processed_disease_drug": [...],109 "processed_disease_targets": [...]110 }111]112```113114---115116## Lower-level functions117118| Function | Input | Output | Description |119|----------|-------|--------|-------------|120| `get_predications(entity, limit=50)` | entity name | `list[dict]` | SemMedDB KG triples |121| `get_ttd_drug_disease(entity)` | drug or disease | `list[dict]` | TTD drug-disease associations |122| `get_ttd_target_disease(entity)` | target or disease | `list[dict]` | TTD target-disease associations |123| `get_processed_drug_disease(entity)` | entity name | `list[dict]` | Processed drug→disease |124| `get_processed_disease_drug(entity)` | entity name | `list[dict]` | Processed disease→drug |125| `get_processed_disease_targets(entity)` | entity name | `list[dict]` | Processed disease→target |126127---128129## Notes130131- All lookups are **case-insensitive** (indexed by lowercased entity names).132- Data is **lazy-loaded**: first `query()` call triggers a one-time index build (may take seconds for large predications file).133- UMLS semantic type codes in predications: `phsu` = pharmaceutical substance, `dsyn` = disease/syndrome, `gngm` = gene/genome, `sosy` = sign/symptom, `podg` = patient/group, etc.134- The GitHub sample `predications.txt` has only 100 lines; for full coverage, download the complete file per the repo README instructions.
gabrielmoreira