Automated Variant Interpretation using LLaMA3 (Ollama)

Overview

This repository demonstrates an automated pipeline for interpreting genetic variants from VCF files.
It combines VEP and ANNOVAR for variant annotation with LLaMA3 via Ollama to generate concise, literature-backed summaries of clinical relevance.

Features

• Annotates variants using VEP and ANNOVAR
• Merges annotation outputs into a single table
• Summarizes the clinical significance of variants using LLaMA3
• Generates CSV reports for downstream analysis or reporting

Repository Structure

llama3-variant-interpretation/
├── data/
│   └── sample_variants.vcf      # Example VCF file with a few variants
├── annotations/
│   ├── vep_output.txt           # Example VEP output
│   └── annovar_output.txt       # Example ANNOVAR output
├── scripts/
│   ├── annotate_variants.py     # Runs VEP + ANNOVAR
│   └── interpret_variants.py    # Merges annotations and calls LLaMA3 via Ollama
├── requirements.txt             # Python dependencies
├── README.md                    # Project overview & instructions
└── .gitignore                   # Ignore unnecessary files

Installation

1. Clone the repository:

git clone https://github.com/<username>/llama3-variant-interpretation.git
cd llama3-variant-interpretation

2. Install Python dependencies:

pip install -r requirements.txt

3. Install VEP and ANNOVAR according to their official documentation:

• VEP
• ANNOVAR

4. Make sure the Ollama server is running locally:

ollama serve

Usage

Step 1: Annotate Variants

Run the annotation script to generate VEP and ANNOVAR outputs:

python scripts/annotate_variants.py

Step 2: Generate Variant Summaries

Run the interpretation script to merge annotations and create literature-backed summaries:

python scripts/interpret_variants.py

Step 3: Check Output

The final summaries will be saved as:

annotations/final_summaries.csv

Sample Data

• data/sample_variants.vcf contains example variants for demonstration.
• Precomputed outputs for VEP and ANNOVAR are included in annotations/ for quick testing.

Notes

• Ensure that the Ollama server is running before executing the interpretation script.
• The pipeline can be extended to batch-process large VCF files.
• LLaMA3 generates summaries based on available annotation data and literature references.

License

This repository is for educational and demonstration purposes. Modify and use according to your needs.