High-Fidelity Data Retrieval from Synthetic DNA Pools via Machine Learning Model.

Area of Science:

• Biotechnology
• Bioinformatics
• Molecular Biology

Background:

• Synthetic DNA offers high information density and stability for data storage.
• Selective data retrieval from complex DNA mixtures remains a key challenge for practical DNA data storage.

Purpose of the Study:

• To develop a machine learning method for high-fidelity, isothermal selective data retrieval from synthetic DNA pools.
• To improve the signal-to-noise ratio for accessing specific data sequences within complex DNA mixtures.

Main Methods:

• Designed a toehold-triggered isothermal DNA storage system with unique stem-loop "lock" sequences for data indexing.
• Trained a machine learning model on a diverse dataset of 12,000 8-nt lock sequences to recognize nucleotide sequence specificity.
• Generated complementary "key" oligos using the trained model to "unlock" specific lock sequences.

Main Results:

• Achieved a maximum improvement of 292-fold in signal-to-noise ratio for selected data sequence amplification.
• Demonstrated high specificity in key oligo design, enabling precise data retrieval.
• The machine learning model learned sequence recognition beyond conventional hybridization principles.

Conclusions:

• The developed machine learning method enables practical, low-energy DNA data storage through isothermal selective retrieval.
• This approach offers insights into DNA sequence specificity, with potential applications beyond data storage.
• High-fidelity retrieval is crucial for the success of DNA-based information storage systems.