Interpretable molecular decision-making with DNA-based scalable and memory-efficient tree computation.

Area of Science:

• Molecular computing
• Biotechnology
• Bioinformatics

Background:

• DNA computing offers molecular-level problem-solving but faces challenges in interpretability, efficiency, and scalability.
• Existing DNA computing methods lack modularity and clear decision-making pathways.

Purpose of the Study:

• To develop a DNA-based decision tree system for enhanced algorithmic interpretability and scalability in molecular computation.
• To demonstrate a modular DNA computing approach for complex classification tasks.

Main Methods:

• Embedding classification rules into DNA strand displacement reaction cascades.
• Constructing cascaded networks exceeding 10 layers and parallel Random Forest computation with 333 DNA strands.
• Integrating a DNA-methylation sensing module for biomarker profile translation.

Main Results:

• Achieved interpretable decision-making across various configurations, including multimode operations (linear/nonlinear, binary/multi-class).
• Demonstrated scalability with cascaded networks and parallel computation of 13 decision trees.
• Enabled accurate disease subtype classification by translating biomarker profiles into molecular instructions.

Conclusions:

• The developed DNA decision tree system offers an interpretable, scalable, and memory-efficient molecular computing solution.
• This approach paves the way for programming intelligent molecular machines with broad applications in diagnostics and beyond.