Prediction of the Near-Infrared Absorption Spectrum of Single-Walled Carbon Nanotubes Using a Bayesian Regularized Back Propagation Neural Network Model

Abstract

DNA-wrapped single-walled carbon nanotubes (DNA-SWCNTs) in stable dispersion are expected to be used as biosensors in the future, because they have the property of absorption of light in the near infrared (NIR) region, which is safe for the human body. However, this practical application requires the understanding of the DNA-SWCNTs’ detailed response characteristics. The purpose of this study is to predict, in detail, the response characteristics of the absorption spectra that result when the antioxidant catechin is added to oxidized DNA-SWCNTs, from a small amount of experimental data. Therefore, in the present study, we predicted the characteristics of the absorption spectra of DNA-SWCNTs using the Bayesian regularization backpropagation neural network (BRBPNN) model. The BRBPNN model was trained with the catechin concentration and initial absorption peaks as inputs and the absorption spectra after catechin addition as outputs. The accuracy of the predicted absorption peaks and wavelengths after the addition of catechin, as predicted by the BRBPNN model, was within 1% of the error of the experimental data. By inputting the catechin concentrations under hundreds of conditions into this BRBPNN model, we were able to obtain detailed prediction curves for the absorption peaks. This method has the potential to help to reduce the experimental costs and improve the efficiency of investigating the properties of high-cost materials such as SWCNTs.

Numerical Analysis for Light Absorption Spectra of the Base of DNA-Wrapped Single-Walled Carbon Nanotubes

“…To solve these problems, material informatics using deep learning may be useful. In research on the light absorption spectra of DNA–SWCNT, a method of creating a Bayesian regularized back-propagation neural network model from a small number of experimental results and predicting various conditions in a data-driven manner has been reported [ 18 ]. Based on the molecular orbital calculation results obtained in this research, the various conditions are expected to be predicted efficiently using a Bayesian regularization back-propagation neural network model based on representative calculation results.…”