Uncertainty Quantification in Machine Learning for Glass Transition Temperature Prediction of Polymers
Code repository for the above titled paper
- Dataset 1: 6097 homopolymers with Tg from PoLyInfo
- Dataset 2: 240 homopolymers with Tg from experiment data
- Dataset 3: 566 homopolymers with Tg from MD simulation
- High-Tg polymers (Tg>350โ): 19 high-Tg polymers from experiment data
- Neural network ensemble: Pytorch
- Gaussian process regression (GPR): GPy
- Monte Carlo dropout (MCD): Pytorch
- Mean-variance estimation (MVE): Pytorch
- Bayesian neural network (BNN): Pytorch
- Evidential deep learning (EDL): Pytorch, Chemprop
- Morgan fingerprint with frequency: Considering the number of substructures
- Mean and standard deviations of Tg for homopolyers
- Spearman's rank correlation coefficient
- Calibration
- Sparsification