Polus: A context-aware enhancement framework for DNA storage via Transformer-based soft-decision decoding

Area of Science:

• Bioinformatics and Computational Biology
• Data Storage Technologies
• Molecular Engineering

Background:

• DNA data storage offers high density and longevity but faces challenges from biochemical noise during synthesis, storage, and sequencing.
• Traditional error correction in DNA storage uses excessive redundancy, compromising efficiency and density.
• Existing methods struggle to effectively mitigate complex error patterns inherent in DNA sequencing.

Purpose of the Study:

• To introduce Polus, a Transformer-based framework designed to enhance the reliability of DNA data storage.
• To improve digital reliability through soft-decision decoding (SDD) without modifying existing encoding schemes.
• To provide a reproducible framework for context-aware decoding and system design guidance in DNA storage.

Main Methods:

• Development of SeqFormer, a Transformer-based channel model that integrates sequence context with quality signals.
• Generation of calibrated per-base confidence scores to convert biochemical noise into informative 'soft' erasures.
• In silico benchmarking of the Polus framework with mainstream DNA storage codecs like DNA Fountain and Yin-Yang.

Main Results:

• Polus significantly improves mainstream DNA storage codecs, reducing sequencing coverage for DNA Fountain by 38.9% and increasing effective physical density by ~80%.
• The framework eliminates persistent indel-induced errors in the Yin-Yang codec and enables targeted resequencing with 99.9% less overhead.
• A comprehensive nine-metric evaluation suite demonstrated multi-dimensional improvements in reliability, density, and cost across DNA storage codecs.

Conclusions:

• Polus provides a novel soft-decision decoding approach that enhances DNA data storage reliability and efficiency.
• The framework effectively transforms biochemical noise into usable information, overcoming limitations of conventional error correction.
• Polus offers a reproducible system for context-aware decoding, guiding future DNA storage system design and optimization.