Predicting glass transition temperature of polyphenol molecular based on its structure

Abstract

Abstract Glass transition is an extremely important phenomenon in condensed physics. It greatly affects the physical property of materials. In this paper, we divide molecular structures into some different types of atom groups according to electrotopological state indices theory. Taking molecular structures and relative molecular mass as variables, we construct models through linear regression and support vector regression (SVR) respectively. Meanwhile we studied how to divide molecular structures into different bonds, considering that the protective degree can directly affect the number of several bonds. Therefore, first we do principle component analysis to eliminate the correlation of the variables by using SVR to construct models. And then, we make comparisons of the two different methods. It is concluded that using the types of atom group to predict the glass transition temperature can achieve higher accuracy.