MOD-KG: MultiOrgan Diagnosis Knowledge Graph

Overview

MOD-KG is a specialized medical Knowledge Graph (KG) designed to capture the complex, interconnected relationships between human organs. It documents how a diagnosis or dysfunction in one organ can influence or cause conditions in others—a phenomenon critical for understanding systemic diseases and multi-organ failure.

The dataset contains over 21,200 knowledge triples, focusing on primary organ systems including the heart, lungs, kidneys, liver, pancreas, and brain, which are central to modern multimodal imaging and clinical research.

Dataset Characteristics

• Scale: 21,200+ unique medical triples (Subject-Relation-Object).
• Evidence-Based: Derived from a hybrid of:
  • Medical Textbooks (13%): Providing foundational physiological relationships.
  • Research Papers: Highly-cited clinical studies (average of 444 citations per paper).
• Methodology: Built using Large Language Models (LLMs) for high-throughput extraction, benchmarked against a 1,000-sample manual baseline to ensure clinical reliability.
• Interconnectivity: Captures "cascading effects" (e.g., how Liver Cirrhosis leads to Diastolic Dysfunction in the Heart).

Repository Structure

The dataset is organized into the following components:

1. final_filtered.csv (The Core Graph)

The primary dataset containing cleaned and unified triples ready for graph analysis or model training.

• Fields: organ1, diagnosis1, relation, organ2, diagnosis2, direction_unified.

2. gpt_extracted.csv (Raw Extractions)

The raw output from the LLM extraction phase, including proof_ids that link claims back to source text.

3. source_analysis.csv (Evidence Metadata)

Detailed metadata for the research papers used, including DOI/Links, citation counts, and journal names.

4. organ.csv (Taxonomy)

A reference file mapping specific diagnoses to their respective organ systems.

5. manual.csv (Benchmark Data)

A manually curated set of triples used to validate the accuracy of the LLM-generated graph.

Usage

This dataset is intended for:

• Bio-medical Informatics: Studying disease propagation across organ systems.
• AI/Machine Learning: Training and fine-tuning Medical LLMs or Graph Neural Networks (GNNs).
• Clinical Decision Support: Predicting secondary complications based on primary organ diagnoses.

Citation

If you use this dataset or paper in your research, please cite:

Anas Anwarul Haq Khan and Pushpak Bhattacharyya. MOD-KG: MultiOrgan Diagnosis Knowledge Graph. In Proceedings of the 2nd Workshop on NLP and AI for Health (NLP-AI4Health) at AACL-IJCNLP 2025.