Machine learning-guided discovery of polymer membranes for CO₂ separation with genetic algorithm

Creators: Basdogan, Yasemin¹; Pollard, Dylan R.; Shastry, Tejus; Carbone, Matthew R.; Kumar, Sanat K.; Wang, Zhen-Gang¹

1. California Institute of Technology

Abstract

Designing polymer membranes with high gas permeability and selectivity is a difficult multi-task constrained problem due to the trade-off between these two properties. In this work, we present a machine learning (ML) driven genetic algorithm to tackle the design problem of polymer membranes for CO separation from N and O Using literature data of permeability for three gases, we constructed multiple ML models with different fingerprinting featurization schemes to predict gas permeabilities. Then, we employed a genetic algorithm to design new polymers and evaluated their performance using our ML models. We were able to identify new polymer membranes that are promising for both CO/N and CO/O separations. The top discovered polymers are predicted to have high glass transition temperatures. Similarly, the pyridine functionality was found in 20% of the predicted polymers. This framework can be used to design polymers for any application involving constrained optimization. Finally, we outlined the challenges and opportunities with using ML guided data-driven inverse design of polymers.

Copyright and License

Acknowledgement

This work was supported in part by Resnick Sustainability Institute (YB, ZGW). Additional support was provided by Hong Kong Quantum AI Lab, AIR@InnoHK of Hong Kong Government (ZGW). DP acknowledges support by the Caltech Amgen Scholars Program. This material is also based upon work supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Award Numbers FWP PS-030 and DE-SC-0012704 (MRC, SKK). At Columbia financial support for this work was provided by the U.S. Department of Energy, United States under Grants DE-SC0021272 (TS).

Supplemental Material

Appendix A. Supplementary data:
Figure S1 shows comparison of ML model predictions with commonly known experimental polymers. Figure S2–S6 shows performance of the GA with different fitness functions. Figure S7 and Table S1 provides Polymer Genome predictions on the hypothetical polymers. Figure S8 shows GA predictions when SAscore of the polymers are included in the fitness function. Figure S9 displays application of ML-driven GA to design O separation membranes.

Appendix B. Supplementary data:
MMC S1.
MMC S2. Initial polymer library as well as the GA generated polymers with permeability for three gasses.

Data Availability

All data used in this research is provided in the SI.

Contributions

Yasemin Basdogan: Writing – original draft, review & editing, Methodology, Funding acquisition, Data curation. Dylan R. Pollard: Writing – review & editing, Methodology. Tejus Shastry: Writing – review & editing, Data curation. Matthew R. Carbone: Writing – review & editing, Formal analysis, Conceptualization. Sanat K. Kumar: Writing – review & editing, Supervision, Formal analysis, Conceptualization. Zhen-Gang Wang: Writing – original draft, review & editing, Supervision, Funding acquisition, Conceptualization.

Files

1-s2.0-S0376738824007634-mmc1.pdf

Name	Size	Download all
1-s2.0-S0376738824007634-mmc2.xls md5:71a732700eaf7591e1b03318e80eb741	4.6 MB	Download
1-s2.0-S0376738824007634-mmc1.pdf md5:74af743a398b4720d50a43c3ff7e692f	8.7 MB	Preview Download

	All versions	This version
Views	0	0
Downloads	0	0
Data volume	0 Bytes	0 Bytes