Imputation of Missing Gas Permeability Data for Polymer Membranes using Machine Learning

Abstract

<p><a>Polymer-based membranes can be used for energy efficient gas separations. Successful exploitation of new materials requires accurate knowledge of the transport properties of all gases of interest. An open source database of such data is of significant benefit to the research community. The Membrane Society of Australasia (https://membrane-australasia.org/) hosts a database for experimentally measured and reported polymer gas permeabilities. However, the database is incomplete, limiting its potential use as a research tool. Here, missing values in the database were filled using machine learning (ML). The ML model was validated against gas permeability measurements that were not recorded in the database. Through imputing the missing data, it is possible to re-analyse historical polymers and look for potential “missed” candidates with promising gas selectivity. In addition, for systems with limited experimental data, ML using sparse features was performed, and we suggest that once the permeability of CO<sub>2</sub> and/or O<sub>2</sub> for a polymer has been measured, most other gas permeabilities and selectivities, including those for CO<sub>2</sub>/CH<sub>4</sub> and CO<sub>2</sub>/N<sub>2</sub>, can be quantitatively estimated. This early insight into the gas permeability of a new system can be used at an initial stage of experimental measurements to rapidly identify polymer membranes worth further investigation.</a></p>

Recommendation System to Predict Missing Adsorption Properties of Nanoporous Materials

“…N.b. recommendation systems have been built for use in the chemical sciences to impute missing gas permeabilities in polymers [71], antiviral activities of molecules [72], and stabilities of inorganic materials [73,74].…”

“…Loosely related, meta-learning has been used to predict an adsorption property of materials at different conditions by learning an intermediate representation of the material based only on available adsorption data. 78 N.b., recommendation systems have been built for use in chemical sciences to impute missing gas permeabilities in polymers, 79 antiviral activities of molecules, 80 and stabilities of inorganic materials. 81,82…”

Discovery of Innovative Polymers for Next-Generation Gas-Separation Membranes using Interpretable Machine Learning

“…The calculated molecular features are then used as inputs for multitask ML models and trained to learn gas permeabilities ( 56 , if at least one gas permeability is available. We use their source code to impute missing gas permeabilities to augment our dataset.…”

“…We utilize two representations for the polymer repeating unit, namely chemical descriptors as generated by RDKit 54 (listed in Table S1 of Supporting Information) and the Morgan fingerprint with frequency (MFF) 55 . We impute missing permeabilities using multivariable imputation by chained equations (MICE) 56 , and we then train multitask supervised ML models to establish synthesis-property relations for these polymer membranes. While various supervised ML models have been used in polymer informatics, including recurrent neural networks, support vector machines, gaussian processes, and others, we choose to focus our study to random forest (RF) regression and deep neural networks (DNN), which have demonstrated outstanding performance in our recent benchmark study 44 .…”

A Recommendation System to Predict Missing Adsorption Properties of Nanoporous Materials

“…Our work is not the first machine-learned recommendation system for use in the chemical sciences. Yuana et al [74] imputed missing gas permeability in polymers. Sosnia et al [75] developed a recommendation system for antiviral drugs by learning a low rank model of a compound-virus activity matrix.…”