Ddi Corpus

DDI Corpus 2013 – Drug-Drug Interaction Query Skill

Overview

The DDI Corpus 2013 is the standard benchmark for drug-drug interaction (DDI) extraction from biomedical text. Each XML file contains sentences with annotated drug entities and pairwise DDI labels.

DDI Types:

• mechanism – pharmacokinetic mechanism described (e.g., altered absorption/metabolism)
• effect – clinical effect of the interaction (e.g., increased bleeding risk)
• advise – recommendation or warning about co-administration
• int – stated interaction without further detail

Entity Types: drug, group, brand, drug_n (active substance not approved for human use)

Setup

1. Download & extract (one-time):

git clone https://github.com/isegura/DDICorpus.git
cd DDICorpus
unzip DDICorpus-2013.zip

2. Set the corpus path in 30_DDI_Corpus_2013.py:

CORPUS_ROOT = "/path/to/DDICorpus-master"  # contains DDICorpus/Train/ and DDICorpus/Test/

Or pass --root at runtime or set env var DDI_CORPUS_ROOT.

Usage

Python API

from 30_DDI_Corpus_2013 import query_entities, list_all_entities, corpus_stats

# Query a single drug
result = query_entities("aspirin")

# Query multiple drugs at once
result = query_entities(["warfarin", "metformin", "digoxin"])

# List all entity names in the corpus
names = list_all_entities()

# Get corpus-level statistics
stats = corpus_stats()

CLI (直接运行)

python 30_DDI_Corpus_2013.py

直接运行即输出 demo 结果（corpus 统计 → 单实体查询 → 批量查询 → 未找到示例）。 修改 __main__ 块中的实体名即可自定义查询。

Output Format

query_entities returns a JSON string. Each element:

{
  "query": "aspirin",
  "found": true,
  "canonical_names": ["ASPIRIN", "Aspirin", "aspirin"],
  "entity_types": ["brand", "drug"],
  "total_interactions": 65,
  "interactions": [
    {
      "partner": "ketoprofen",
      "ddi_type": "mechanism",
      "sentence": "concurrent administration of aspirin decreased ketoprofen protein binding...",
      "source": "Train/DrugBank"
    }
  ],
  "example_sentences": ["..."]
}

If an entity is not found: {"query": "xyz", "found": false}.

Parameters

Dependencies

Python 3.10+ standard library only (xml.etree.ElementTree, json, os, collections). No third-party packages required.