This repository contains a local, research-oriented knowledge graph for colorectal carcinoma. It is rooted at the NCI Thesaurus (NCIt) concept C2955 / Colorectal Carcinoma and connects the disease taxonomy to explicitly related NCIt concepts, including genes, molecular abnormalities, anatomical sites, pathological findings, stages, and treatment regimens.
The repository also provides a local incremental workflow for extracting concepts and relations from new materials, reviewing candidate entities, producing a separate updated graph, and answering graph-grounded questions with optional PubMed and LLM support.
这是一个面向结直肠癌研究的本地知识图谱系统。图谱以 NCIt C2955 / Colorectal Carcinoma 为根，保留其疾病概念树，并纳入通过 NCIt role 或 association 明确关联的基因、分子异常、解剖部位、细胞/组织、临床 finding、分期和治疗方案。
系统坚持以下数据模型：概念和实体是节点；NCIt 原始关系是边；同义词、定义、UMLS CUI、ICD-O-3、来源和证据句是属性。它用于术语整理、知识浏览、资料增量整理、研究问答和假设生成，不用于替代个体化临床诊疗判断。
.
├─ extract_ncit_colorectal.py # 从 NCIt EVSREST API 获取 C2955 概念树
├─ enrich_related_ncit_entities.py # 补全树外关联 NCIt 对象及其同义词
├─ build_colorectal_kg.py # 生成主图谱、HTML、GraphML、Cytoscape JSON
├─ ncit_colorectal_cancer/ # NCIt 原始/中间数据快照
├─ colorectal_knowledge_graph/ # 主知识图谱产物
├─ kg_update_system/ # 新资料抽取、合并、问答与本地 GUI
├─ scripts/validate_release.py # 发布数据完整性检查
├─ DATA_PROVENANCE.md # 数据来源、版本与边界
├─ CITATION.cff # 引用信息
├─ CODE_OF_CONDUCT.md # 协作行为规范
└─ SECURITY.md # 安全与密钥处理规则
| Item | Value |
|---|---|
| Source terminology | NCI Thesaurus (NCIt) |
| API source | NCI EVSREST API |
| Root concept | C2955 / Colorectal Carcinoma |
| NCIt version | 26.04d |
| Snapshot generated | 2026-05-14 |
| Disease-tree concepts | 285 |
| Related external NCIt entities | 463 |
| Total graph nodes | 748 |
Detailed provenance and semantic boundaries are documented in DATA_PROVENANCE.md.
The main README is bilingual. Chinese counterparts are available for the operational documents below:
| English | 中文版 |
|---|---|
| Contributing | 贡献指南 |
| Data Provenance and Scope | 数据来源与范围 |
| Security Policy | 安全策略 |
| Code of Conduct | 协作行为规范 |
| Changelog | 版本记录 |
LICENSE and CITATION.cff retain their standard international formats. Their scope and use are explained in the Chinese sections of this README and the Chinese data-provenance document.
Core graph construction uses Python's standard library.
python .\build_colorectal_kg.py
python .\kg_update_system\kg_gui.pyOpen http://127.0.0.1:8765 when the GUI starts. The generated main graph can also be opened directly from:
colorectal_knowledge_graph/kg_browser.html
colorectal_knowledge_graph/kg_tree.html
Run the release-integrity check:
python .\scripts\validate_release.pyThe bundled data are a versioned snapshot. To refresh it from NCIt, run the following commands in order:
python .\extract_ncit_colorectal.py
python .\enrich_related_ncit_entities.py
python .\build_colorectal_kg.pyThis operation accesses the official NCIt EVSREST API and may change node counts and relationships when NCIt releases a new version.
python .\kg_update_system\kg_incremental.py extract `
--input .\kg_update_system\sample_materials `
--kg-dir .\colorectal_knowledge_graph `
--output .\kg_update_system\runs\sample_mini
python .\kg_update_system\kg_incremental.py merge `
--base-kg .\colorectal_knowledge_graph `
--mini .\kg_update_system\runs\sample_mini `
--output .\kg_update_system\runs\updated_kgThe merge command writes a new graph directory and does not overwrite colorectal_knowledge_graph. New candidate entities require explicit review and --accept-candidates before inclusion.
Copy the safe template before enabling LLM-enhanced extraction or answers:
Copy-Item .\kg_update_system\config.example.json .\kg_update_system\config.deepseek.json
$env:KG_LLM_API_KEY="your-provider-key"The local AI configuration is intentionally excluded from Git. Do not commit API keys, clinical records, or unredacted sensitive materials.
If this repository supports your work, cite the software metadata in CITATION.cff and cite NCIt as the upstream terminology source. See DATA_PROVENANCE.md for the exact snapshot details.
Contribution conventions are described in CONTRIBUTING.md. Please also follow CODE_OF_CONDUCT.md and avoid submitting patient-identifiable data, API keys, or unredacted clinical materials.
- NCIt role and association relations are terminology assertions, not patient-specific clinical recommendations.
Disease_May_Have_*describes a possible relationship; it does not mean every patient has that feature.- LLM-extracted candidate entities and relations require human review against source evidence.
- The repository contains no patient-level data; users must not add identifiable medical information to a public fork.
Project code is released under the MIT License. Upstream NCIt data retain their own provenance and usage conditions; the repository license does not supersede those conditions.