Recommended by
Indexed by
Publication date: Apr 27, 2022
The computational generation of molecules with Artificial Intelligence (AI) is poised to revolutionize materials discovery. Potential applications range from development of potent drugs to efficient carbon capture and separation technologies. However, existing approaches lack either automatized training data creation or computational performance validation at meso-scale where complex properties of amorphous materials emerge. The methodological gaps have so far limited AI based materials design to small-molecule applications. Here, we report the first automatized discovery of complex materials through inverse molecular design which is informed by meso-scale target features resembling application-level figures-of-merit. We have entered the new discovery regime by computationally designing and validating hundreds of unknown polymer candidates for application in post-combustion carbon dioxide separation. Specifically, we have validated each discovery step, from automatized input data set creation, via graph-based generative design of optimized monomer units, to molecular dynamics simulation of gas filtration by newly discovered polymer membranes. For the latter, we have devised a Minimum Representative Volume method enabling reliable polymer permeability predictions at about 1,000x the volume of an individual, AI-generated monomer. The overall AI discovery and physical validation cycle time per polymer candidate is of the order of 200 hours in a standard computing environment, offering an alternative to lab screening routines that can take months to complete. The inclusion of additional target properties in the optimization workflow and the extension of generative algorithms to the design of larger, more complex molecular entities will improve the outcome and benefit the broader applicability of computational materials discovery. The data-set show the CO2 permeability, glass transition temperature and half decomposition temperature for 1,200 homo-polymers used to train our AI model.
No Explore or Discover sections associated with this archive record.
| File name | Size | Description |
|---|---|---|
|
ppf_API_demo.ipynb
MD5md5:468d276cf957d2636c5876ac76f9f973
|
11.1 KiB | Jupyter notebook with an example of how to use the Polymer Property Prediction Engine with APIs (application programming interface) |
|
example.csv
MD5md5:77a1abfca19c1a53bfc954c0372338b6
|
550 Bytes | Input file data-set for the Jupyter notebook |
|
PCO2-Tg-Thd-data-all-simulated.csv
MD5md5:534cf22962db6838cb93cd023a1c2306
|
189.1 KiB | 1,169 calculated homo-polymers properties: IUPAC polymer name, OPSIN SMILES, log10( CO2 permeability (Barrer)), glass transition temperature (K) and half decomposition temperature (K) |
|
results_monomers_generative_model.csv
MD5md5:50e094dcc18de52b03a67cc6d4e8a6c9
|
81.7 KiB | 784 homopolymers generated by Inverse Design and Machine Learning techniques with the OPSIN SMILES, glass transition temperature (Tg), half decomposition temperature (Thf) and CO2 permeability (PCO2). Tg and Thf are calculated using Polymer Property Prediction Engine, and PCO2 was calculated by the Machine Learning regression model. |
| 2023.20 (version v7) | Jan 30, 2023 | DOI10.24435/materialscloud:64-c8 |
| 2023.14 (version v6) | Jan 25, 2023 | DOI10.24435/materialscloud:ma-eq |
| 2022.99 (version v5) | Jul 27, 2022 | DOI10.24435/materialscloud:ma-qn |
| 2022.65 (version v4) | May 18, 2022 | DOI10.24435/materialscloud:jk-zm |
| 2022.61 (version v3) | May 10, 2022 | DOI10.24435/materialscloud:p8-ey |
| 2022.58 (version v2) | Apr 28, 2022 | DOI10.24435/materialscloud:zm-hb |
| 2022.56 (version v1) [This version] | Apr 27, 2022 | DOI10.24435/materialscloud:ss-fq |