An artificial intelligence-aided virtual screening recipe for two-dimensional materials discovery

Abstract

In recent years, artificial intelligence (AI) methods have prominently proven their use in solving complex problems. Across science and engineering disciplines, the data-driven approach has become the fourth and newest paradigm. It is the burgeoning of findable, accessible, interoperable, and reusable (FAIR) data generated by the first three paradigms of experiment, theory, and simulation that has enabled the application of AI methods for the scientific discovery and engineering of compounds and materials. Here, we introduce a recipe for a data-driven strategy to speed up the virtual screening of two-dimensional (2D) materials and to accelerate the discovery of new candidates with targeted physical and chemical properties. As a proof of concept, we generate new 2D candidate materials covering an extremely large compositional space, downselect 316,505 likely stable 2D materials, and predict the key physical properties of these new 2D candidates. Finally, we hone in on the most propitious candidates of functional 2D materials for energy conversion and storage.

Crystal-Site-Based Artificial Neural Networks for Material Classification

“…The elements of this one-hot vector were zeros and ones, depending on the number of atoms in the material formula. Similar non-structural approaches have been reported by Jha et al [36] and Goodall et al [37] Embeddings of elemental properties have been perhaps the most concurred material characterization approach [3,[38][39][40][41], which allows the combination of properties of different magnitudes of scale and nature. Isayev and coworkers [40] characterized the materials for the AFLOW repository with the so-called property-labeled materials fragments.…”

“…In recent years, machine learning algorithms have irrupted as an alternative tool to model the properties and structure of materials [1][2][3][4][5][6][7][8][9][10][11]. These algorithms have allowed scientists to work with large particle systems at shorter times and lower computational costs with respect to the recurred quantum methods [12][13][14][15].…”

Data-Driven Discovery of 2D Materials for Solar Water Splitting

“…While Court et al (2020) used a somewhat similar approach in their model, lack of an explicit representation of the lattice precluded the application of their model to non-orthogonal systems. Furthermore, all these generative models hold an advantage over high-throughput virtual screening approaches such as those reported by Sorkun et al (2020), since they possess the capability to not only identify new material compositions but also new phases for known material compositions. However, this in no way undermines the importance of high-throughput screening approaches.…”

“…In addition, machine learning models have also been trained using data from these repositories to predict properties of novel materials (Ahmad et al, 2018;Xie and Grossman, 2018;Ye et al, 2018;Joshi et al, 2019;Liu et al, 2020). In a recent article, Sorkun et al (2020) identified several potential 2D materials for photocatalytic water splitting, CO 2 reduction, and N 2 reduction by training AI models on the computational 2D materials database and using the predictions from these models to screen a vast chemical space obtained by systematic elemental substitution in 2D material prototypes.…”

Atomic-scale one-dimensional materials identified from graph theory

“…As compared with conventional nanomaterials (nanosheets, nanofibers and quantum dots) that may possess unsaturated dangling bonds, the atomic-scale LDMs are characterized by fully terminated surfaces, which are chemically inert and thus confine the intrinsic properties and functionalities of the LDMs 10,11 . In recent years, with the aid of state-of-the-art high-throughput first principles calculations, a multitude of atomic-scale LDM candidates have been discovered, predominantly in the form of twodimensional (2D) atomic layers [12][13][14][15][16][17][18][19] . Most of them are either identified via top-down exfoliation of existing bulk materials, or anticipated by bottom-up chemical substitutions of already-synthesized 2D compounds [16][17][18] .…”