🧬 DNA-AI: Private Genetic Analyzer

DNA-AI is a local, privacy-focused bioinformatics tool that analyzes raw DNA data (from 23andMe, AncestryDNA, etc.) against the NCBI ClinVar database. It combines deterministic data matching with a local Large Language Model (Llama 3) to explain health risks in plain English.

⚠️ DISCLAIMER: This tool is for educational and research purposes only. It is NOT a medical device. Always consult a genetic counselor or doctor for medical advice.

🚀 Features

• 100% Private & Offline: Your DNA data never leaves your computer. The AI runs locally.
• Robust Matching: Matches user DNA against the ClinVar database using Chromosome and Position.
• Smart Filtering:
  • Zygosity Detection: Distinguishes between Carriers (1 Copy) and Affected (2 Copies).
  • Strand Flip Protection: Automatically detects and hides ~90% of false positives caused by reverse-strand sequencing errors (e.g., A>T or C>G palindromes).
  • Strict Mode: Hides variants with "Conflicting interpretations" among scientists.
• AI Geneticist: Chat with a local Llama 3 AI to ask questions about your specific results.
• Universal Loader: Handles .txt, .csv, .zip, and .gz files automatically.

---

🛠️ Prerequisites

Before running the program, ensure you have the following installed:

1. Python 3.10+
2. Ollama: (Required for the AI features). Download here.

1. Setup the AI Model

Open your terminal/command prompt and run:

ollama pull llama3

2. Install Python Libraries

It is recommended to use a virtual environment.

# Create venv
python -m venv venv

# Activate venv (Windows)
venv\Scripts\activate

# Install requirements
pip install streamlit pandas langchain langchain-community langchain-chroma langchain-ollama pypdf

---

📂 Data Setup

You need two files to run this program.

1. The Database (ClinVar)

You need the "Variant Summary" file from NCBI.

• Download Link: variant_summary.txt.gz
• Location: NCBI FTP Site
• Note: Do NOT unzip this file. The program reads the .gz directly to save space.

2. Your DNA Data

• Download your "Raw Data" from 23andMe, AncestryDNA, or MyHeritage.
• The file should look like a text list of rsid, chromosome, position, genotype.

---

▶️ How to Run

1. Open your terminal in the project folder.
2. Activate your virtual environment (if used).
3. Run the Streamlit app:

streamlit run dna_chat_app.py --server.maxUploadSize=2000

---

🔍 How to Interpret Results

The Filters

• ⚠️ Confirmed Risks Only: Hides "Benign" and "Uncertain" results.
• 🔥 Strict Mode: Hides results where labs disagree (e.g., one lab says "Pathogenic" but another says "Benign").
• 🧬 Hide Strand Ambiguity: (Recommended: ON). Hides ambiguous "Palindrome" mutations (A↔T, C↔G) which are often technical errors in 23andMe data.

Zygosity (The most important column)

• Heterozygous (1 Copy / Carrier): You have one normal gene and one mutated gene. For recessive conditions, you are usually healthy but can pass it on.
• Homozygous (2 Copies) ⚠️: You have two mutated genes. This is a significant finding and warrants further investigation or discussion with a professional.

---

🐛 Troubleshooting

"Found 0 matches"

• Ensure your DNA file uses Build 37 (GRCh37/hg19) coordinates (standard for 23andMe/Ancestry).
• If using very new clinical data (Build 38), the positions will not match ClinVar.

"Found 30,000 matches"

• This is normal before filtering. Turn on "Confirmed Risks Only" and "Hide Strand Ambiguity".

"The AI isn't responding"

• Make sure Ollama is running in the background. Open a separate terminal and type ollama serve.

---

📜 License

This project is open-source. Data provided by NCBI ClinVar. Public domain test data provided by Harvard Personal Genome Project.