Inverse Design of Polymers via Evolutionary Algorithm and Bayesian optimization
We propose a robust machine learning workflow for the inverse design of high thermal conductivity (TC) polymers. Our work starts from a computational database containing 1144 polymers with known TCs. Using those data, we construct a surrogate deep neural network model for TC calculation and extract a polymer-unit library with 32 sequences. Two state-of-the-art algorithms of unified non-dominated sorting genetic algorithm III (U-NSGA-III) and q-noisy expected hypervolume improvement (qNEHVI) are employed for sequence-controlled polymer design. They are the multi-objective evolutionary algorithm and multi-objective Bayesian optimization algorithm, respectively, since the synthesizability of the emerging polymers is also evaluated using the synthetic accessibility score. The approach proposed is flexible and universal, and can be extended to the design of polymers with other property targets. Please refer to our work "AI-assisted inverse design of sequence-ordered high intrinsic thermal conductivity polymers" for additional details.
To download, clone this repository:
git clone https://github.com/SJTU-MI/Inverse_Design_of_Polymers.git
To run most code in this repository, the relevant anaconda environment can be installed from environment.yml. To build this environment, run：
cd ./Inverse_Design_of_Polymers
conda env create -f environment.yml
conda activate IDPoly
Cal_TC.py: A script for calculating thermal conductivity of polymers, here predicted by a DNN surrogate model
Cal_SA.py: A script for evaluating the synthetic accessibility scores of polymers
utility.py: Some utility functions
MOEA_candidates.csv: Polymers designed by MOEA
MOEA_HV.csv: Convergence of the MOEA based on hypervolumes
MOBO_candidates.csv: Polymers designed by MOBO
MOBO_HV.csv: Convergence of the MOBO based on hypervolumes
MOEA_Case.ipynb: A case for MOEA
MOBO_Case.ipynb: A case for MOBO
visualization.ipynb: Visualization of results for MOEA and MOBO cases
| AUTHORS | Xiang Huang, Shenghong Ju |
|---|---|
| VERSION | V1.0 / October,2023 |
| EMAILS | [email protected] |
- pymoo(Multi-objective Optimization in Python)[Link]
- botorch(Bayesian optimization in PyTorch)[Link]
- RadonPy(automated physical property calculation using all-atom classical molecular dynamics simulations for polymer informatic)[Link]
- Huang, X., Zhao, C.Y., Wang, H., et al., AI-assisted inverse design of sequence-ordered high intrinsic thermal conductivity polymers[J]. Materials Today Physics. 2024: 101438. [Link].
- Huang, X., Ju, S., Tutorial: AI-assisted exploration and active design of polymers with high intrinsic thermal conductivity[J]. Journal of Applied Physics. 2024, 135 (17):171101. [Link].
This work is under BSD-2-Clause License. Please, acknowledge use of this work with the appropiate citation to the repository and research article.