Machine learning-directed massively parallel programmable nucleic acid amplification.

Area of Science:

• Molecular Biology
• Biotechnology
• Bioinformatics

Background:

• Precise control over nucleic acid amplification efficiency is crucial for molecular diagnostics and DNA data storage applications.
• Existing methods face challenges in dynamic regulation and achieving high resolution in amplification control.

Purpose of the Study:

• To develop a thermodynamics-based approach for continuous and precise modulation of nucleic acid amplification efficiency.
• To enhance DNA data storage capabilities and improve sensitivity in molecular diagnostics.

Main Methods:

• Implemented a primer-tag compensation strategy to decouple sequence specificity from hybridization energy regulation.
• Developed a machine learning model using 2483 experimental data points to predict amplification efficiency, improving accuracy from R²=0.62 to R²=0.86.
• Validated the amplification strategy in DNA data storage and clinical applications, including cervical cancer RNA variant analysis.

Main Results:

• Achieved programmed amplification with high resolution (33% vs 81% efficiency).
• Increased information preview density in DNA data storage by nearly one order of magnitude.
• Demonstrated a 100-fold improvement in detection sensitivity for rare RNA fusions in cervical cancer analysis.

Conclusions:

• The developed programmable amplification technique offers precise control over efficiency, benefiting DNA data storage and molecular diagnostics.
• This method significantly enhances detection sensitivity for rare molecular targets and has potential applications in single-cell sequencing and spatial transcriptomics.