rnaseq-local-chat

Chat with your RNA-seq results using a fully local LLM — no API keys required.

Upload DESeq2 tables, GSEA pathway results, sample metadata, and figures (volcano plots, UMAPs) and ask questions about them in a NotebookLM-style interface.

Requirements

• Ollama installed and running
• llama3.2 model: ollama pull llama3.2
• llava model (for image understanding): ollama pull llava

Setup

pip install -r requirements.txt

Usage

streamlit run app.py

Then open http://localhost:8501 in your browser.

1. Upload files in the sidebar
2. Click Build / Update DB
3. Ask questions in the chat

Or use the CLI ingestion tool:

python ingest.py --files data/deseq2_results.csv data/volcano.png --db ./chroma_db

Supported File Types

File	Naming Convention	What is stored
DESeq2 results	contains deseq, deg, differential	Per-gene: log2FC, padj, baseMean, direction
GSEA results	contains gsea, pathway, kegg, hallmark	Per-pathway: NES, padj, direction, size
Sample metadata	contains meta, sample, coldata	Per-sample conditions + summary
Images (PNG/JPG)	any filename	LLaVA-generated text description

Example Questions

• "Which genes are significantly upregulated (padj < 0.05, log2FC > 1)?"
• "What immune-related pathways are activated?"
• "What does the volcano plot show?"
• "How many samples are in each condition?"
• "Which cancer hallmark pathways are enriched?"

Project Structure

rnaseq-local-chat/
├── app.py          # Streamlit UI
├── ingest.py       # CSV + image → ChromaDB (importable + CLI)
├── data/           # Put your RNA-seq files here
├── requirements.txt
├── .env.example
└── LICENSE

Stack

• LLM: llama3.2 via Ollama
• Vision: llava for image-to-text
• Embeddings: all-MiniLM-L6-v2
• Vector store: ChromaDB
• Framework: LangChain LCEL
• UI: Streamlit