Molecular implementation of a DNA-based artificial neural network

Molecular implementation of a DNA-based artificial neural network

Molecular convolutional neural networks with DNA regulatory circuits. (a) Left: ConvNet architecture. Right: Diagram of the ConvNet operating principle for recognition tasks. (B) An unknown input pattern is added to the solution, when a DNA circuit is added, the fluorescence signals can be read to report the identification results. (c) DNA-ConvNet can recognize eight Chinese oracles. Credit: Xiong et al.

Molecular computing is a promising field of study that aims to use biological molecules to create programmable devices. This idea was first introduced in the mid-1990s and has since been realized by many computer scientists and physicists around the world.

Researchers at East China Normal University and Shanghai Jiao Tong University recently developed molecular convolutions neural networks (CNNs) based on synthetic DNA regulatory circuits. Their approach, it was presented in a paper published in The intelligence of nature’s machineovercomes some of the challenges we typically encounter when creating efficient artificial neural networks based on molecular components.

“The intersection of computer science and molecular biology It is fertile ground for exciting new science, especially since intelligent systems design is a long-term goal for scientists,” Hao Bai, one of the researchers who conducted the study, told TechXplore. Compared to the brain, scale and computing, the power of sophisticated DNA neural networks is very limited, due to network size restrictions. The primary goal of our work was to increase the computing power of DNA circuits by offering a suitable solution neural network A model of molecular DNA systems”.

While conducting their research, Bai and colleagues found that it could be particularly promising for modeling DNA circuits, given their sparse topological connectivity, which is similar to real biological neural networks. So they decided to use CNNs to create a large-scale DNA-based neural network consisting of 512 synthetic DNA strands. Notably, their proposed network produces thousands of chimical interaction It generates hundreds of molecular species.

“Our group focused on the micro-engineering and programming of nucleic acid molecules, and we designed and built a series of dynamic DNA nanostructures that can be used as regulatory elements to build large circuits,” Bay explained. “In this work, we use a dynamic DNA nanostructure called a switch gate, which is functionally similar to riboswitches in genes. Regulatory departmentsall consist of two independent functional areas that sense and respond to external inputs.

The switching gate in the network allows researchers to independently control signal transduction and weight assignment functions through a process known as intramolecule conformational switching. This process is particularly suitable for weight sharing and sparse connectivity of CNNs.

DNA circuits operate within the researchers’ network, where all computing units are ready to respond to input. Once the input is fed into the solution, the single DNA strands will trigger successive strand displacement reactions in order.

These interactions, which are driven by Gibbs free energy or entropy in the system, generate the corresponding fluorescent signals. The researchers encoded all test patterns using a set of individual DNA strands and each of the fluorescence signals generated represented one such test pattern.

“We extend the main advantage of the sparse CNN topology and weight-sharing architecture into a DNA neural network, which can effectively reduce the complexity and parameters of Network Engineering Through few contact neurons. To implement these features, we designed a switch gate geometry composed of two independent functional domains (the weight tuning domain and the recognition domain). With this design, we can easily change the sequence design of the corresponding functional domain successively to suit the tuning of the network architecture. “

The CNN-based approach has several advantages over the previously proposed molecular computing methods. First, its switch-gate architecture can be used to embed ligand-responsive molecular switches. This would allow the network to adapt its functions in response to environmental changespotentially enabling the development of molecular circuits resembling biological neural networks and capable of ‘intelligent’ behaviour.

In addition, the inherently parallel nature of DNA molecules could enable independent parallelization of CNN processes. This can be particularly useful for achieving scalable information processing.

“We have proposed a systematic strategy for implementing the ConvNet algorithm at the molecular level,” said Bay. “We feel our method is a major advance in synthetic molecular information processing systems, achieving fast and accurate classification tasks that can classify 32 molecular patterns within 30 minutes, which may be the fastest and most powerful synthetic chemical computing system to date, to the best of our knowledge.”

Recent work by Pai and colleagues provides an alternative DNA-based architecture that could aid in the design of new molecular computing systems. In the future, their approach could be used to create several molecular diagnostic devices for biomedical applications.

“By linking sensory inputs, a DNA-based ConvNet could in principle use hundreds of targets as inputs and facilitate broader applications in disease diagnosis, expression patterns, and precision medicine,” Bay added. “Based on the DNA-based ConvNet model, we now plan to build a molecular classifier that can be used to classify multiple disease diagnoses.”


LOEN: a lensless electronic neural network enables device vision


more information:
Xiewei Xiong et al., Molecular convolutional neural networks with DNA regulatory circuits, The intelligence of nature’s machine (2022). DOI: 10.1038 / s42256-022-00502-7

Wei Lai et al, Programming chemical reaction networks using DNA molecular conformational motions, ACS nano (2018). DOI: 10.1021 / acsnano.8b02864

Wei Lai et al, Nonlinear organization of enzyme-free DNA circuits with ultrasensitive switches, ACS Synthetic Biology (2019). DOI: 10.1021 / acssynbio.9b00208

© 2022 Science X Network

the quote: Molecular Implementation of a DNA-Based Artificial Neural Network (2022, Jul 25) Retrieved Jul 25, 2022 from https://techxplore.com/news/2022-07-molecular-dna-based-artustry-neural-network.html

This document is subject to copyright. Notwithstanding any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.