🧬 Gene Variant Pathogenicity Predictor

An analysis application that allows users to simulate single-nucleotide variations (SNVs) in genes and predict whether they are “likely benign” or “likely pathogenic”, using the Evo2 LLM Model.

---

🚀 Overview

This project consists of:

• A modified Evo2 LLM tokenizer for efficient inference with SNVs.
• A Next.js frontend to display gene information, allow input of nucleotide changes, and view predictions.
• Deployment to a Modal container, bundling the Evo2 model with custom tweaks.

---

🧪 Functionality

• View gene details and reference sequences.
• Introduce a single-nucleotide variation in the sequence.
• Predict the impact of that mutation using the Evo2 LLM.
• Classify the result as:
  • Likely Benign
  • Likely Pathogenic

---

🛠️ Modifications

After cloning the original Evo2 Repository, a small change was made in the tokenizer when trying to run in the Modal Container (It is recommended to use H100 GPUs):

# File: vortex/model/tokenizer.py (under CharLevelTokenizer)

def tokenize(self, text: str):
    return list(np.frombuffer(text.encode("utf-8"), dtype=np.uint8))

The updated Evo2 directory is added to the Modal container with (as mentioned under main.py):

.add_local_dir("evo2", remote_path="/evo2", ignore=["*.venv", "*.ipynb"], copy=True)

🧱 Tech Stack

• Next.js with ShadCN – frontend UI for sequence interaction

• Python (with Modal) – model backend for Evo2 inference

• Evo2 LLM – large language model for protein/nucleotide analysis

• NumPy, Pandas, FastAPI, Matplotlib – backend utility libraries

• Scikit-learn - for evaluation metrics

📦 Setup & Usage

1. Clone and install dependencies (from their respective directories)

   cd backend/
   pip install -r requirements.txt

   cd frontend/
   npm install

2. Modify the Evo2 tokenizer (already included if using this repo)

3. Run locally (Modal / Backend):

   cd backend/
   modal init 
   modal run main.py

4. Start the Next.js frontend:

   cd frontend/
   npm run dev