首页|Polymer informatics based on the quantitative structure-property relationship using a machine-learning framework for the physical properties of polymers in the ATHAS data bank
Polymer informatics based on the quantitative structure-property relationship using a machine-learning framework for the physical properties of polymers in the ATHAS data bank
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NSTL
Elsevier
In polymer informatics based on the quantitative structure-property relationship (QSPR) through machine learning (ML), one of the key issues is how to utilize a high-quality database of polymer properties. The ATHAS data bank is one of the valuable databases of polymer-specific physical properties, including glass transition temperature (T-g) and heat capacity difference at T-g for fully amorphous polymers, equilibrium melting point and heat of fusion at 100% crystallinity, etc. Using the ATHAS data bank, QSPRs between fingerprints of repeating polymeric structural units and each physical property were obtained by ML. Two types of hidden-layer structures of artificial neural network (ANN) were examined as regression models, and their optimal hidden-layer structures were determined by the contour plots of root mean square error in each physical property. In both ANN structures, a good correlation was found between the registered values and the predicted ones, suggesting that the physical properties may be predicted from only the repeating polymeric structural units. Furthermore, the physical properties of poly(p-dioxanone), which are not yet registered in the ATHAS data bank, were predicted, indicating that the predicted properties agreed with the measured properties from literature within +/- 25% in the practical temperature range. A new method for predicting heat capacity for polymers is proposed by combining ML and ATHAS analysis.
NSTL
Elsevier
