Biologically Constrained DNA Encoding With Triplet Networks for Similarity Image Retrieval

Area of Science:

• Bioinformatics
• Data Storage
• Computer Science

Background:

• Exponential growth of digital data necessitates novel storage solutions.
• DNA offers high density and durability for long-term data archiving.
• Efficient encoding strategies are crucial for DNA data retrieval.

Purpose of the Study:

• To develop a DNA encoder training framework for improved accuracy and efficiency in content-based image retrieval.
• To incorporate deep metric learning into DNA encoding.
• To enforce biological constraints for enhanced DNA sequence stability.

Main Methods:

• Proposed a training framework for a DNA encoder using deep metric learning.
• Introduced novel loss functions to control homopolymer length and GC content.
• Evaluated performance using image classification on CIFAR-10 and CIFAR-100 datasets.

Main Results:

• Achieved classification accuracy comparable to CNN-based baselines.
• Demonstrated a 20-fold speedup in training time compared to existing methods.
• Ensured strict control of homopolymer length and optimal GC content (40-60%) for biological feasibility.

Conclusions:

• The proposed method enhances DNA data storage for image retrieval with improved accuracy and training efficiency.
• Incorporating biological constraints improves the biochemical stability of DNA sequences.
• The approach offers a more biologically feasible DNA data storage solution.