TransDNA: A Deep Transfer Learning Network for Sequence Reconstruction in DNA-Based Data Storage.

Area of Science:

• Biotechnology
• Computer Science
• Data Storage

Background:

• DNA data storage offers high density and durability but faces challenges in accurate information recovery due to errors.
• Sequence reconstruction is critical for decoding DNA data, yet limited training samples hinder deep learning approaches.

Purpose of the Study:

• To develop an effective deep learning method for DNA sequence reconstruction that overcomes the limitation of scarce training data.
• To introduce TransDNA, a deep transfer learning network designed to improve the accuracy and efficiency of DNA data recovery.

Main Methods:

• Proposed TransDNA, a deep transfer learning network comprising an encoder, domain-specific decoder, and domain-invariant feature extractor.
• Employed alternating domain alignment and domain-specific reconstruction mechanisms.
• Utilized knowledge transfer from a larger source dataset to enhance performance on smaller target datasets from real DNA storage experiments.

Main Results:

• TransDNA significantly improved the sequence reconstruction success rate on real DNA storage datasets.
• The proposed method outperformed a base model without transfer learning and other comparative methods.
• TransDNA demonstrated superior performance compared to the SDG method in both reconstruction success rate and training efficiency.

Conclusions:

• TransDNA is the first successful application of transfer learning to the DNA sequence reconstruction task.
• The developed method effectively addresses the challenge of limited training data in DNA storage systems.
• TransDNA offers a promising solution for enhancing the reliability of information recovery in DNA data storage.