Nonlinear decision-making with enzymatic neural networks.

Area of Science:

• Biotechnology
• Molecular computing
• Synthetic biology

Background:

• Artificial neural networks have transformed electronic computing.
• Molecular networks offer potential for biological decision-making comparable to gene regulatory networks.
• Previous non-enzymatic neuromorphic architectures faced limitations in sensitivity, speed, and nonlinear response.

Purpose of the Study:

• To introduce DNA-encoded enzymatic neurons with tunable properties.
• To construct multilayer neuromorphic architectures for molecular classification.
• To achieve classification of nonlinearly separable regions in molecular data.

Main Methods:

• Utilizing DNA-encoded enzymatic neurons with adjustable weights and biases.
• Assembling neurons into multilayer networks for complex computations.
• Developing hybrid circuits combining neural and logical operations within cell-sized droplets.

Main Results:

• Demonstrated computation of majority functions on 10-bit inputs using individual neurons.
• Constructed a two-layer network to synthesize rectangular functions based on microRNA input.
• Created a hybrid circuit that recursively partitions a concentration plane using a decision tree.

Conclusions:

• DNA-encoded enzymatic neurons enable multilayer architectures for classifying nonlinearly separable molecular data.
• This approach offers computational power and miniaturization for analyzing complex molecular systems.
• Potential applications include querying liquid biopsies and DNA databases.