yotamnahum / DNA-Data-Storage

Single Read Reconstruction for DNA Data Storage Using Transformers (official implementation)

Single Read Reconstruction for DNA Data Storage Using Transformers

Overview

This repository is the official implementation of "Single Read Reconstruction for DNA Data Storage Using Transformers", a novel approach leveraging an encoder-decoder Transformer architecture for DNA-based data storage. In the face of increasing global data storage demands, DNA storage emerges as a promising solution with high information density. However, DNA synthesis and sequencing are prone to errors. Our model uniquely addresses this by facilitating single-read reconstruction, significantly reducing reading costs and achieving lower error rates compared to traditional multi-read methods.

Key Features

• Transformer Architecture: A groundbreaking deep learning model specifically designed for error correction in DNA sequences.
• Single-Read Reconstruction: First-of-its-kind approach using deep learning for single-read reconstruction in DNA-based storage, significantly reducing overall costs.
• Versatility in Data Types: Capable of handling and reconstructing diverse data types, including text, images, and code scripts.
• Efficient and Accurate: Demonstrates lower error rates in reconstruction from single DNA strand reads than existing state-of-the-art algorithms.

Installation

Run pip install -r requirements.txt to install required packages.

Usage

• Configuration: Customize model configurations in configuration.py.
• Data Processing: Employ data_utils for efficient data loading and preparation.
• Model Training: Initiate model training through DNAmodel.py.
• Diverse Data Handling: Utilize image_utils.py and text_utils.py for conversion and reconstruction of images and text data.
• Prediction and Evaluation: Execute predictions and evaluate performance using Predict.py.

Components

• configuration.py: Settings and parameters for the model.
• data_utils: Modules for data handling and preprocessing.
• DNAmodel.py: Core deep learning model and associated methods.
• image_utils.py, text_utils.py: Tools for processing image and text data.
• Predict.py: Functions for model predictions and evaluations.
• utils.py: General utilities, including high-fidelity error simulation methods.

Requirements

Python packages required are listed in requirements.txt.

Citation

If you find this work useful, please cite our paper:

@misc{nahum2021singleread,
      title={Single-Read Reconstruction for DNA Data Storage Using Transformers}, 
      author={Yotam Nahum and Eyal Ben-Tolila and Leon Anavy},
      year={2021},
      eprint={2109.05478},
      archivePrefix={arXiv},
      primaryClass={cs.ET}
}

License

This project is licensed under the MIT License.

For more details, refer to the full paper: Single-Read Reconstruction for DNA Data Storage Using Transformers.