Uncertainty Quantification in Machine Learning for Polymer Properties: A Benchmark Study

---

Code repository for the above titled paper. In notebook, we provide a step-by-step guide to reproduce the results of Tm in the paper.

Workflow

Dataset

• Tg: Glass transition temperature; Eg: Band gap; Tm: Melting temperature; Td: Decomposition temperature.
• Tg^EXP: represents out-of-distribution experimental Tg data from journal papers.
• Tg^MD: represents out-of-distribution Tg data derived from MD simulations in the PoLyInfo.

Methods

• 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

Input

• Morgan fingerprint with frequency (MFF): Considering the number of substructures

Output

• Mean and standard deviations for polymer properties

Metrics

• R2, MAE, RMSE
• Spearman's rank correlation coefficient
• Calibration

More details can be found in the paper.