Pattern recognition in the nucleation kinetics of non-equilibrium self-assembly

Area of Science:

• Biophysics
• Computational Science
• Materials Science

Background:

• Neural networks, inspired by the brain, are advanced computational architectures.
• Similar high-dimensional, interconnected systems exist in cellular processes like signal transduction and genetic networks.
• The study explores if analogous collective modes exist in non-information-processing physical and chemical systems.

Purpose of the Study:

• To investigate if nucleation during multicomponent self-assembly exhibits information processing capabilities.
• To determine if high-dimensional concentration patterns can be discriminated and classified like neural networks.
• To design and test a DNA tile system for classifying image data.

Main Methods:

• Designed 917 DNA tiles for self-assembly into three distinct structures.
• Utilized in silico training to classify 18 grayscale images into three categories based on nucleation patterns.
• Employed fluorescence and atomic force microscopy for experimental validation over 150 hours.

Main Results:

• The DNA tile system successfully classified all trained images.
• Experimental results confirmed the computational classification accuracy.
• Variations in the test set demonstrated the robustness of the nucleation-based classification.

Conclusions:

• Nucleation in multicomponent self-assembly can perform computations analogous to neural networks.
• This physical phenomenon possesses inherent information-processing capabilities.
• The developed DNA-based system is compact, robust, scalable, and offers a novel approach to computation.