Applying Machine Learning and Quantum Chemistry to Predict the Glass Transition Temperatures of Polymers
Description
Developing models to predict glass transition temperatures (Tgs) of polymers is of significant importance given how the parameter quantifies the physical and thermal characteristics of these materials. These characteristics inform numerous functional properties of polymers as well as how they degrade both through intended use as well as through environmental mechanisms that result from end-of-life environmental deposition. For this reason, various models have been developed for predicting Tg from structural information to aid in designing novel polymer materials. These existing models, however, typically focus on utilizing one specific modeling technique to train a single class or set of polymers. To expand and explore the applicability of Tg models for predictions of new materials this work (1) utilizes both machine learning (ML) and quantum chemistry (QC) based techniques to investigate different data availability scenarios and (2) does not constrain the training datasets to any specific class of polymers. This methodology allows for a comparison of different techniques and situations to determine the applicability of Tg models when making predictions for novel polymer structures.
Files
Steps to reproduce
Experimental glass transition temperatures (Tgs) for both models were pulled from the literature as described in the paper. The ML_Model directory contains all of the raw data and python scripts necessary to reproduce the model results with a graph convolutional network, as well as the scripts necessary for plotting the results. The QC_Model directory contains the raw data for Tgs as well as input files for the Gaussian 16 calculations and their subsequent output files. Also included are the R scripts necessary to read the output files and construct the final stepwise model, as well as the final script for running and plotting the results.