Online Constructive Machine Learning with Molecular Hypernetworks in DNA Computing

Various benefits of DNA computing such as programmability, immense data storage capacity and massively parallel processing have led to provide fundamental building blocks to motivated goals of building smart in vivo robots with implications in various applications. However, molecular machine learning has not yet been employed to solve more complex tasks such as pattern recognition, due to the lack of control of molecules in liquid state, instability and inaccuracy of in vitro manipulation to carry out learning algorithms experimentally. Here, a potential link between DNA computing and machine learning is proposed through the complementary action of a constructive machine learning model, the hypernetwork for DNA computation. Constructive DNA self-assembly and enzymatic weight update underlies the experimental scheme for wet lab implementation and online constructive learning of hypernetworks. We introduce both theoretical and practical details of the proposed model, present preliminary experimental results for examining the practical viability and discuss further applications.