Prediction of polymer properties using infinite chain descriptors (ICD) and machine learning: Toward optimized dielectric polymeric materials

To facilitate the development of new polymeric materials, we report the development of new heuristic models to predict the dielectric constant, band gap, dielectric loss tangent, and glass transition temperatures for organic polymers. A new set of features called infinite chain descriptors (ICDs) was designed and developed especially to characterize organic polymers, utilizing methods with minimal dependence on pre-defined fragment libraries. Machine learning models were built for the aforementioned properties incorporating best practices in the field such as objective feature selection, cross-validation and external test sets. All models produced in this study showed good performance in prediction. A web tool has been developed and has been made available that supports the input of novel structures. (C) 2016 Wiley Periodicals, Inc.