Ananth Govind Rajan’s group has developed a novel string-based representation for nanopores in 2D materials, which is human interpretable and machine learnable. Its application will significantly advance the design of nanoporous materials for separation applications.
Nanoporous 2D materials, like graphene, possess unique properties that make them ideal for applications such as gas separation, seawater filtration, and even DNA sequencing. The size and shape of the nanopores primarily determine these properties. However, the wide variability in nanopore geometries poses significant challenges in predicting and controlling the performance of these materials. Previous approaches, such as mathematical combinatorics of polyforms to enumerate all possible nanopore shapes and graph-based methods to compare shapes, have provided some insights. However, these methods are computationally intensive and lack the capability for reverse engineering—predicting nanopore shapes and sizes with specific desired properties. To overcome these limitations, we developed a novel framework called STring Representation Of Nanopore Geometry (STRONG). STRONG is a language that encodes the shape and structure of nanopores into a sequence of characters, making it both human-readable and machine-interpretable. This language assigns unique symbols to different atomic configurations along the nanopore edges, enabling precise representation of nanopore geometries.
Using STRONG, we established structure-property relationships for nanoporous 2D materials. We trained a recurrent neural network (RNN), a machine learning model commonly used in natural language processing, to “read” this encoded language and predict nanopore properties such as formation energy and gas transport barriers. This approach accelerates property predictions and lays the foundation for reverse engineering nanopores with tailored properties. Looking ahead, the STRONG framework holds significant potential for creating digital twins of 2D materials. By integrating experimental data, researchers can use STRONG to reconstruct the nanopore distributions responsible for a material’s observed properties. This capability could revolutionize the design and optimization of nanoporous materials for a wide range of applications, making STRONG a valuable tool for scientists and engineers alike.
The Application images are adapted from open-access articles (Appl. Sci. 2018, 8(9), 1547, npj 2D Mater. Appl. 2021, 5, 66, and ACS Appl. Mater. Interfaces 2017, 9(1), 92).
