ekacare/BODHI-M

---
license: cc-by-nc-4.0
language:
- en
tags:
- medical
- knowledge-graph
- clinical
- healthcare
- india
- snomed
- loinc
- graph
- drug
- lab-investigation
- medication
pretty_name: BODHI-M — Clinical Concept-Drug-Lab Investigation Knowledge Graph
size_categories:
- 1K<n<10K
---

# BODHI-M — Concept-Drug-Lab Investigation Knowledge Graph

Part of **BODHI (Bharat Ontology for Disease & Healthcare Informatics)** — open clinical knowledge graphs for grounding healthcare AI in verified medical facts.

→ [Full writeup: motivation, design & use cases](https://info.eka.care/services/bodhi-bharat-ontology-for-disease-healthcare-informatics)
→ [GitHub (all formats: Neo4j, CSV, PyG, RDF)](https://github.com/eka-care/BODHI)

---

## What is BODHI-M?

BODHI-M maps SNOMED-coded clinical concepts (disorders, findings, procedures, lifestyle factors) to their generic drug treatments and LOINC-coded lab investigations, organised in a three-level hierarchy: **System → Group → Granular**.

Built and validated by expert clinicians at [Eka Care](https://www.eka.care), it has powered production patient health profiling and longitudinal health views across millions of records in India.

The three-tier concept hierarchy is a core design choice: when a drug or lab result cannot confidently pinpoint a specific granular disease, the graph supports reliable "soft inference" at the broader System or Group level. This also enables **reverse inference** — deducing likely conditions from a patient's medication list alone.

## Stats

| Metric | Count |
|---|---|
| Concept nodes | 2,471 |
| Drug nodes | 1,186 |
| LabInvestigation nodes | 812 |
| **Total relationships** | **3,566** |
| Concept → Concept (CHILD_OF) | 1,768 |
| Concept → Drug (TREATED_BY) | 908 |
| LabInvestigation → Concept (IMPACTS) | 808 |
| Concept → LabInvestigation (MONITORED_BY) | 82 |

**Concept hierarchy:** System `14` → Group `250` → Granular `1,942`

**LabInvestigation LOINC coverage:** 812 LOINC-mapped tests across Immunological, Renal, Hematological, Endocrine, and Gastrointestinal domains.

## Files

| File | Description |
|---|---|
| `triples.jsonl` | `(head, relation, tail, properties)` structured triples |
| `nl_facts.jsonl` | Natural-language fact strings, suitable for LLM fine-tuning / RAG |

For Neo4j dump, CSV, PyTorch Geometric, and RDF/Turtle formats, see the [GitHub repository](https://github.com/eka-care/BODHI).

## Schema (triples)

Each line in `triples.jsonl`:
```json
{
"head": "<node_id>",
"head_type": "Concept | Drug | LabInvestigation",
"relation": "CHILD_OF | TREATED_BY | IMPACTS | MONITORED_BY",
"tail": "<node_id>",
"tail_type": "Concept | Drug | LabInvestigation",
"properties": { ... }
}
```

## Standards

- **SNOMED CT** — all concept nodes carry SNOMED IDs
- **LOINC** — all lab investigation nodes carry LOINC IDs

## Use Cases

- **Reverse inference** — deduce likely conditions from a patient's medication history
- **Patient health profiling** — build richer longitudinal views from fragmented health data
- **GraphRAG** — structured grounding for LLMs on treatment and investigation reasoning
- **GNN training** — heterogeneous graph with multi-class nodes and typed edges

## License

[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) — free for non-commercial use with attribution to [Eka Care](https://www.eka.care).

许可证：CC BY-NC 4.0
语言：
- 英语（en）
标签：
- 医学（medical）
- 知识图谱（knowledge graph）
- 临床（clinical）
- 医疗保健（healthcare）
- 印度（india）
- SNOMED
- LOINC
- 图谱（graph）
- 药物（drug）
- 实验室检查（lab investigation）
展示名称：BODHI-M — 临床概念-药物-实验室检查知识图谱（Clinical Concept-Drug-Lab Investigation Knowledge Graph）
规模类别：
- 1K<n<10K

# BODHI-M — 临床概念-药物-实验室检查知识图谱

本数据集属于**BODHI（疾病与医疗信息学印度本体，Bharat Ontology for Disease & Healthcare Informatics）**系列，是旨在将医疗人工智能（AI）锚定在经过验证的医学事实之上的开源临床知识图谱集。

→ [完整文档：研究动机、设计方案与应用场景](https://info.eka.care/services/bodhi-bharat-ontology-for-disease-healthcare-informatics)
→ [GitHub仓库（包含Neo4j、CSV、PyG、RDF等全格式文件）](https://github.com/eka-care/BODHI)

## 什么是BODHI-M？

BODHI-M将经过SNOMED编码的临床概念（包括疾病、体征、操作、生活方式因素）映射至对应的通用药物治疗方案与经过LOINC编码的实验室检查项目，并采用**系统→类别→细粒度**的三级层级结构进行组织。

本数据集由[Eka Care](https://www.eka.care)的临床专家构建并验证，已在印度境内支撑了数百万条医疗记录的患者健康画像与纵向健康视图生产任务。

三级概念层级是本数据集的核心设计思路：当药物或实验室检查结果无法精准定位某一具体细粒度疾病时，该知识图谱可在更宽泛的系统或类别层面实现可靠的“软推理”。同时，这一设计也支持**反向推理**——仅通过患者的用药清单即可推导其可能罹患的疾病。

## 统计数据

| 指标 | 数量 |
|---|---|
| 概念节点 | 2,471 |
| 药物节点 | 1,186 |
| 实验室检查节点 | 812 |
| **总关系数** | **3,566** |
| 概念→概念（CHILD_OF） | 1,768 |
| 概念→药物（TREATED_BY） | 908 |
| 实验室检查→概念（IMPACTS） | 808 |
| 概念→实验室检查（MONITORED_BY） | 82 |

**概念层级结构**：系统级 `14` → 类别级 `250` → 细粒度级 `1,942`

**实验室检查LOINC覆盖范围**：覆盖免疫学、肾脏学、血液学、内分泌学与胃肠病学领域的812项经LOINC编码的检查项目。

## 文件说明

| 文件 | 说明 |
|---|---|
| `triples.jsonl` | 采用`(head, relation, tail, properties)`结构的结构化三元组文件 |
| `nl_facts.jsonl` | 自然语言事实文本，适用于大语言模型（Large Language Model, LLM）微调与检索增强生成（RAG） |

如需获取Neo4j导出文件、CSV、PyTorch Geometric以及RDF/Turtle格式的数据，请参阅[GitHub仓库](https://github.com/eka-care/BODHI)。

## 三元组数据格式

`triples.jsonl`中的每一行格式如下：
json
{
"head": "<node_id>",
"head_type": "Concept | Drug | LabInvestigation",
"relation": "CHILD_OF | TREATED_BY | IMPACTS | MONITORED_BY",
"tail": "<node_id>",
"tail_type": "Concept | Drug | LabInvestigation",
"properties": { ... }
}

## 遵循标准

- **SNOMED CT**：所有概念节点均附带SNOMED编码
- **LOINC**：所有实验室检查节点均附带LOINC编码

## 应用场景

- **反向推理**：仅通过患者的用药史推导其可能罹患的疾病
- **患者健康画像**：从碎片化的医疗数据中构建更完整的纵向健康视图
- **图检索增强生成（GraphRAG）**：为大语言模型（LLM）在治疗与检查推理方面提供结构化锚定基础
- **图神经网络（Graph Neural Network, GNN）训练**：包含多类别节点与带类型边的异构图数据集

## 许可证

[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)：可免费用于非商业用途，但需注明引用[Eka Care](https://www.eka.care)。