ragbio 0.1.13

RAG pipeline for gene-disease literature summarization

RAG-Powered Gene Discovery Assistant (ragbio)

A generative AI tool for biomedical knowledge discovery using Hugging Face embeddings and Ollama LLMs (DeepSeek / LLaMA3). This project integrates retrieval-augmented generation (RAG) with PubMed literature and gene annotation data to summarize gene–disease relationships. Now packaged as a reusable Python package, it can be imported and used in multiple bioinformatics projects.

---

Overview

The RAG-powered assistant enables:

• Semantic search over PubMed abstracts and gene annotations.
• Summarization of complex biomedical information.
• Citation tracking with PubMed IDs.
• Modular and reusable pipeline for gene–disease exploration.
• Case study–specific queries using query_name to organize outputs and visualizations.

Example queries:

• "Which genes are linked to oxidative stress in Alzheimer’s disease?"
• "Summarize recent findings about TP53 variants in cancer."

---

Architecture

User Query
│
▼
Vector Retrieval (BioBERT / BioSentVec Embeddings)
│
▼
Top Abstracts + Gene Annotations
│
▼
Ollama LLM (DeepSeek / LLaMA3)
│
▼
Summarized Biomedical Answer + Citations
│
▼
Optional: Gene–Disease–Drug Network (Cytoscape / Streamlit)

---

Installation

Clone and install as a package:

git clone https://github.com/<your-username>/rag-gene-discovery-assistant.git
cd rag-gene-discovery-assistant
pip install -e .

Example requirements.txt

langchain
faiss-cpu
sentence-transformers
biopython
pymed
requests
sqlite-utils
ollama
pandas
beautifulsoup4
streamlit

---

Usage

1. Fetch PubMed Data

from ragbio.utils.rag_data_loader import main as fetch_pubmed_data

# Download abstracts and metadata
fetch_pubmed_data()

2. Run RAG Query (with query_name)

from ragbio.pipeline.rag_pipeline import RAGAssistant

# Use query_name to organize output
assistant = RAGAssistant(output_dir="output/Alzheimer_CaseStudy", query_name="Alzheimer_CaseStudy")
summary, pmids, structured = assistant.run_pipeline(
    "genes linked to Alzheimer’s disease",
    top_k=10,
    structured=True
)

Outputs are stored under output/<query_name>/ for easier tracking.

---

3. Visualize Gene–Disease–Drug Networks

You can visualize the structured outputs in Cytoscape via Streamlit:

streamlit run ragbio/pipeline/rag_cytoscape_streamlit.py -- --query_name Alzheimer_CaseStudy

This reads all JSON files from output/<query_name>/ and plots the gene–target–drug–disease network.

Example Output Network

Figure 1: Gene–disease–drug co-occurrence network generated from top PubMed abstracts.

---

4. Optional: Explore in Notebook

Open notebooks/RAG_GeneDiscovery_Assistant.ipynb to see example queries, visualizations, and outputs.

---

Technologies Used

Category	Tool
Embeddings	BioBERT, BioSentVec (Hugging Face)
LLM Backend	DeepSeek / LLaMA3 (Ollama)
Retrieval	FAISS
Data Sources	PubMed, UniProt, NCBI Gene
Language	Python 3.10+
Frameworks	LangChain, Sentence Transformers, Streamlit

---

Future Enhancements

• Compare DeepSeek/LLaMA3 with BioGPT outputs.
• Integrate Neo4j for gene–disease–drug knowledge graph visualization.
• Fine-tune LLMs on curated variant interpretation reports for improved clinical relevance.
• Extend package API for direct integration in Django, FastAPI, and Streamlit apps.
• Add support for multiple query_name outputs to track different case studies.