Machine Learning Systems Design using Molecular and DNA Reactions

Abstract

There is a growing recognition that machine learning can play an important role in a wide range of critical applications, such as data mining, natural language processing, and biomedical applications. Design and synthesis of molecular machine learning systems are of interest as these have the potential to revolutionize applications such as in-situ protein monitoring, drug delivery, and molecular therapy. In modern practice, a disease is diagnosed by collecting data from a body sensor, analyzing the data in a computer or laboratory to diagnose a disease, and then delivering a therapy for either prevention or cure. In the proposed molecular biomedicine framework, the sensing, analytics, feature computation, and therapy would all be in the same place, i.e., in-vivo. In this dissertation, we synthesize different machine learning systems through molecular reactions, where inputs and outputs are chemical molecules, e.g., DNA strands. First, we propose a novel approach to synthesize molecular reactions to compute a radial basis function (RBF) support vector machine (SVM) kernel. The approach is based on fractional coding where a variable is represented by two molecules. The synergy between fractional coding in molecular computing and stochastic logic implementations in electronic computing is key to translating known stochastic logic circuits to molecular computing. We introduce a new explicit bipolar-to-unipolar molecular converter for intermediate format conversion. Two designs are presented; one is based on the explicit and the other is based on an implicit conversion from prior stochastic logic. When five support vectors are used, it is shown that the DNA RBF-SVM realized using the explicit format conversion has orders of magnitude less regression error than that based on implicit conversion. Second, we present an innovative method for synthesizing molecular reactions with the aim of training a perceptron, i.e., a single-layered neural network, with a sigmoidal activation function. The approach is also based on fractional coding. A new molecular scaler that performs multiplication by a factor greater than 1 is proposed based on fractional coding. The training of the perceptron proposed is based on a modified backpropagation equation as the exact equation cannot be easily mapped to molecular reactions using fractional coding. Third, we present the implementation of artificial neural networks (ANNs) using molecular computing and DNA based on fractional coding. Prior work had addressed molecular two-layer ANNs with binary inputs and arbitrary weights. We make four contributions. First, molecular perceptrons that can handle arbitrary weights and can compute sigmoid of the weighted sums are presented. Thus, these molecular perceptrons are ideal for regression applications and multi-layer ANNs. A new molecular divider is introduced and is used to compute sigmoid(ax) where a > 1. Second, based on fractional coding, a molecular artificial neural network (ANN) with one hidden layer is presented. Third, a trained ANN classifier with one hidden layer from seizure prediction application from electroencephalogram is mapped to molecular reactions and DNA and their performances are presented. Fourth, molecular activation functions for rectified linear unit (ReLU) and softmax are also presented. We then present novel implementations for reservoir computing (RC) using DNA oscillators. An RC system consists of two parts: reservoir and readout layer. The reservoir projects input signals into a high-dimensional feature space which is formed by the state of the reservoir. The internal connectivity structure of the reservoir remains unchanged. After training, the readout layer maps the projected features into the desired output. It has been shown in prior work that coupled deoxyribozyme oscillators can be used as the reservoir. We utilize the n-phase molecular oscillator (n >= 3). The readout layer implements a matrix-vector multiplication using molecular reactions based on molecular analog multiplication. All molecular reactions are mapped to DNA strand displacement (DSD) reactions. We also introduce a novel encoding method that can significantly reduce the reaction time. The feasibility of the proposed RC systems based on the DNA oscillator is demonstrated for the handwritten digit recognition task and a second-order nonlinear prediction task. Finally, we propose molecular and DNA memristors where the state is defined by a single output variable. Past molecular and DNA memristors defined the state of the memristor based on two output variables. The prior memristors cannot be cascaded because their input and output sizes are different. We introduce a different definition of state for the molecular and DNA memristors. This change allows cascading of memristors. The proposed memristors are used to build reservoir computing (RC) models. We also study the input-state characteristics of the cascaded memristors and show that the cascaded memristors retain the memristive behavior. The cascade connections in a reservoir can change dynamically; this allows the synthesis of a dynamic reservoir as opposed to a static one in the prior work. This reduces the number of memristors significantly compared to a static reservoir. The inputs to the readout layer correspond to one molecule per state instead of two; this significantly reduces the number of molecular and DNA reactions for the readout layer. A DNA RC system consisting of DNA memristors and a DNA readout layer can be used to solve the seizure detection task. We also demonstrate that a DNA RC system consisting of three cascaded DNA memristors and a DNA readout layer can be used to solve the time-series prediction task.