General Network Learning Rules Based on DNA Strand Displacement for Thyroid Disease Prediction.

Abstract

Learning rules are critical to the problem-solving ability of neural networks. Significant progress has been made in neural networks based on deoxyribonucleic acid (DNA) strand displacement (DSD). Traditional chemical reaction networks (CRNs) usually focus on the implementation of one type of learning rule. The coexistence of multiple learning rules remains challenging. In this article, CRNs based on DSD are constructed. The networks consist of a weight multiplication module, an activation function module, a learning signal module, a weight update module, and a weight output module. By exploring the concentration of auxiliary strands in the modules, discrete perceptron, Hebbian, and filtered learning rules are simulated successfully. The feasibility is verified through a simple instance. Modules are also used to build a classification model that can learn about thyroid disease and make predictions about test categories. The simulation is verified by the software Visual DSD. This article will provide a theoretical basis for biomedical prediction and identification.