A systematic approach to generate polymer library: A search for polymers with high dipole moment

Abstract

Four machine learning (ML) models are trained using seven molecular descriptors for the prediction of dipole moment. Random forest model has stand out as best model with r-squared values of 0.82 and 0.67 for training and test set, respectively. 10 k polymers are generated using automated process and dipole moment of newly generated polymers is calculated using random forest model. A significant change in dipole moment on structural change is observed. Polymers are screened on the basis on their dipole moment values. Majority of chosen polymers are easy to synthesize. The chosen polymers have revealed resemblances among their structures.