Two of the trendiest terms in science - big data and machine learning have been brought together to help a team based at Columbia Engineering in New York City and the University of South Carolina in the USA, and the Max Planck Society in Mainz, Germany, develop a new approach to designing semi-permeable polymer membranes that can filter gases. Ultimately, such membranes might be used to scrub carbon dioxide from waste gas and so reduce carbon emissions.

Details are published in the journal Science Advances and represent the first example of an experimentally validated and rapid machine learning approach to polymer design for gas separation membranes. "Our work points to a new way of materials design and we expect it to revolutionize the field," explains study leader Sanat Kumar.

Polymer membranes are widely used to separate gas mixtures, allowing carbon dioxide, nitrogen, and methane to be filtered from gaseous mixtures. Advanced membranes might be useful in removing carbon dioxide to purify natural gas (methane) or to separate it for the purposes of carbon capture. Of course there are hundreds if not tens of thousands of chemically plausible polymers that might be investigated to this end. So far researchers have fabricated just a thousand or so, each at great cost for little return. A high-throughput process that can screen hundreds of possible structures quickly could revolutionize the field, the team suggests.

A machine learning algorithm can quickly correlate the chemical structure of hundreds of known polymers for their gas transport properties and then use this "learning" to screen thousands more to find an optimal polymer for the job. Indeed, the team has already identified about 100 polymers that had not previously been tested for gas transport properties. They have found these to all be predicted to outperform any current membrane used for carbon dioxide and methane separation.

"Rather than experimentally test all the materials that exist for a particular application, you instead test a smaller subset of materials which have the most promise. You then find the materials that combine the very best ingredients and that gives you a shot at designing a better material," explains co-author Connor Bilchak. The performance of two of the cohort of polymers made in the laboratory for testing outshone any known polymer membrane materials. A similar approach might be applied to other types of polymer membrane, such as those that might be used to build the next-generation of rechargeable battery or for more effective water purification. [Barnett, J.W. et al. Sci Adv (2020); 6 (20): eaaz4301 DOI: 10.1126/sciadv.aaz4301]