$\text{Uni}D^3$: Open Knowledge Graphs and QA Benchmarks for Drug–Disease Discovery and Reasoning

📄 Paper: UniD³: A knowledge graph-enhanced RAG framework for drug-disease discovery and reasoning (arXiv:2606.01394)

💬 Web Chatbot: https://unid3.site/

📦 Knowledge graphs & vector database (Zenodo): https://zenodo.org/records/15368180

Overview

Drug--disease knowledge underpins drug discovery and precision medicine, yet most of it is locked in unstructured PubMed articles and only partially captured by curated databases. We release UniD3 -KG, a suite of six knowledge graphs spanning three tasks (Drug--Disease Matching, DDM; Drug Effectiveness Assessment, DEA; and Drug--Target Analysis, DTA), and UniD3 -data, three open QA datasets ( DDM, DEA and DTA pairs), distilled from PubMed articles with Llama-3.3-70B by the UniD3 KG-RAG pipeline, which couples task-specific prompting with a dual-stage entity extraction that re-extracts over the consolidated graph to recover cross-chunk and cross-document evidence. External validation on CDR, DrugReviews, P3ps and MedicationQA and a benchmark across seven LLM baselines and three retrieval variants show that UniD3 -data is faithful to curated sources and that retrieval over UniD3 -KG consistently improves biomedical QA; a blinded clinician review of FDA-approved, US-marketed drugs further confirms strong agreement with the UniD3 -data DDM split (accuracy , F1 , AUROC ). All artefacts---UniD3 -KG, UniD3 -data, benchmark scripts and the UniD3 -QA explainable chatbot---are released as shared infrastructure for biomedical NLP and drug repurposing. The interactive UniD3 web chatbot is now live at https://unid3.site/, and the generated KGs and vector database are available on Zenodo at https://zenodo.org/records/15368180.

UniD3 QA （under LightRAG framework（required lightrag-hku==1.1.0））

# Please download the generated knowledge graph and vector database from Zenodo (https://zenodo.org/records/15368180) before using UniD3 QA. 
# Please specify the specific working path, large language model and RAG mode.
$ conda env create -f UniD3_environment.yaml
# run ollama before here
$ python QA_launcher.py --working_dir "/UniD3/KG_building_level2/level2_T2_70B" --model "myllama3.3_70B" --mode "mix"

Dataset Usage

The three QA datasets (DDM, DEA, DTA) are publicly available on the Hugging Face Hub:

Load them directly with the 🤗 datasets library:

from datasets import load_dataset

ddm = load_dataset("Mike2481/UniD3_DDM")
dea = load_dataset("Mike2481/UniD3_DEA")
dta = load_dataset("Mike2481/UniD3_DTA")

Resources

Citation

If you find this project is useful for your research, please cite:

Wang Q, Liu T, Zhou M, Liang J, Guo S, Wang G, Su J, Song Q. UniD³: A knowledge graph-enhanced RAG framework for drug-disease discovery and reasoning. arXiv. 2026; arXiv:2606.01394. https://arxiv.org/abs/2606.01394

License

MIT License (anonymized for review; copyright information will be added upon paper acceptance).