Machine learning with persistent homology and chemical word embeddings improves prediction accuracy and interpretability in metal-organic frameworks

Abstract

Machine learning has emerged as a powerful approach in materials discovery. Its major challenge is selecting features that create interpretable representations of materials, useful across multiple prediction tasks. We introduce an end-to-end machine learning model that automatically generates descriptors that capture a complex representation of a material’s structure and chemistry. This approach builds on computational topology techniques (namely, persistent homology) and word embeddings from natural language processing. It automatically encapsulates geometric and chemical information directly from the material system. We demonstrate our approach on multiple nanoporous metal–organic framework datasets by predicting methane and carbon dioxide adsorption across different conditions. Our results show considerable improvement in both accuracy and transferability across targets compared to models constructed from the commonly-used, manually-curated features, consistently achieving an average 25–30% decrease in root-mean-squared-deviation and an average increase of 40–50% in R2 scores. A key advantage of our approach is interpretability: Our model identifies the pores that correlate best to adsorption at different pressures, which contributes to understanding atomic-level structure–property relationships for materials design.

Chemistry-Encoded Convolutional Neural Networks for Predicting Gaseous Adsorption in Porous Materials

“…If only MOFs that are potentially suitable for separation and storage applications (i.e., k H,CH4 > 10 –7 [mol/kg/Pa]) are considered, the RMSE value is even lower to be 0.18. Such a small RMSE value suggests the suitability of the CNN model in a high-throughput screening and it is also among the best machine learning models that have been reported to date for methane adsorption under dilute conditions to the best of our knowledge …”

“…While classical molecular simulations represent an important means toward the search of optimal materials, conducting an exhaustive computational screening in a brute-force manner is still resource-demanding. To this end, machine learning (ML) has recently drawn considerable attention for its great potential in discoveries of high-throughput materials. − For example, Cai’s group employed support vector machine (SVM) and tree-based regressions with 37 tailor-made features to study 130 397 hypothetical MOF structures and shed light on the optimal combinations of structural features (e.g., void fraction and surface area) for methane storage . Woo and co-workers also implemented SVM-based machine learning to predict MOFs’ methane deliverable capacity and elucidated the quantitative structure–property relationship (QSPR) .…”

Transport-Relevant Pore Limiting Diameter for Molecular Separations in Metal–Organic Framework Membranes

“…To this end, we envision that PLD, particularly PLD t‑r proposed in this study, can still serve as a very useful means of preliminary screening of materials. Machine learning (ML) may also be a promising approach to improve the prediction accuracy, as it has been successfully employed for predicting the adsorption properties of various gases in MOFs. − Besides PLD t‑r , a variety of other easy-to-compute features such as LCD, void fraction, and open-metal-site angle can be used as input features to train an ML model for the prediction of gaseous transport properties in porous materials.…”

Hydrogen storage in MOFs: Machine learning for finding a needle in a haystack

“…To address this challenge, the use of ML in conjunction with MOFs has also grown significantly in the past 5 years, from gas adsorption prediction to partial atomic charges, band gap, and other mechanical and chemical property predictions. 2 , 3 , 4 , 5 , 6 , 7 Due to their porous nature, fast kinetics, reversibility, and high gravimetric densities, MOFs have been widely studied for many gas-storage problems, including natural gas and hydrogen. Hydrogen storage is a key enabling technology as hydrogen is considered both a future automotive fuel and a medium for energy storage; however, its application has been limited by hydrogen’s low volumetric density at ambient conditions.…”