Science Undergraduate Laboratory Internship Project

Predicting Polymer Properties via Graph Convolutional Network

Graph representations of molecules have been used to accurately predict the properties of polymers, in some cases with more success than typical feature-list molecular representations. The purpose of this project is to recreate the paper published by Park et al, Prediction and Interpretation of Polymer Properties Using the Graph Convolutional Network, in order to affirm that graph representations of molecules can be used to predict polymer properties. By means of recreating the Park study, the relationship between thermal and mechanical properties and structural characteristics of monomer units is explored by implementing a graph convolutional network (GCN) to model and predict the glass transition temperature, melting temperature, and density for polymers curated from the PolyInfo open-access database. This project focuses on building a comparable graph convolutional network machine learning model that can reproduce the accuracy of the Park model and extended to applications beyond the scope of the Park study. Additionally, improvements for Park's machine learning model are proposed in this project.​