Machine‐Learning Prediction of Metal–Organic Framework Guest Accessibility from Linker and Metal Chemistry

Abstract

The choice of metal and linker together define the structure and therefore the guest accessibility of a metal‐organic framework (MOF), but the large number of possible metal‐linker combinations makes the selection of components for synthesis challenging. We predict the guest accessibility of a MOF with 80.5 % certainty based solely on the identity of these two components as chosen by the experimentalist, by decomposing reported experimental three‐dimensional MOF structures in the Cambridge Structural Database into metal and linker and then learning the connection between the components’ chemistry and the MOF porosity. Pore dimensions of the guest‐accessible space are classified into four ranges with three sequential models. Both the dataset and the predictive models are available to download and offer simple guidance in prioritization of the choice of the components for exploratory MOF synthesis for separation and catalysis based on guest accessibility considerations.

MOFGalaxyNet: a social network analysis for predicting guest accessibility in metal–organic frameworks utilizing graph convolutional networks

“…We extracted the linker SMILES string notation, metal ion, and pore limiting diameter (PLD) data label from the curated CSD [ 11 ]. The metal atomic number, weight, radius, Milliken electronegativity, polarizability, and electron affinity for the metal ion were calculated using the freely available software Mordred [ 34 ].…”

“…First, we considered the node classification problem in MOFGalaxyNet, where the labels defined as PLD are only available for a subset of MOFs. Therefore, the continuous value of PLD must be adapted to four ranges defined in the previous study [ 11 ]. The four ranges are defined as nonporous (PLD < 2.4 Å), small pores (2.4 Å < PLD < 4.4 Å), medium pores (4.4 Å < PLD < 5.9 Å), and large pores (5.9 Å < PLD).…”

“…Consequently, models that predict the properties of MOFs by using only this information about the constituents as input will be more suitable for a first narrowing down of the MOF chemical spaces containing structures suited for a desired application. In a recent study [ 11 ], a new ML model was designed to predict the guest accessibility of MOFs using only information about the linkers and the metal ions. Here, guest accessibility was defined as the diameter of the largest free sphere that a guest molecule can diffuse through the MOF, also known as the pore-limiting diameter (PLD).…”

Unified roadmap for ZIF-8 Nucleation and Growth: Machine Learning Analysis of Synthetic Variables and their Impact on Particle Size and Morphology

“…12,[14][15][16] It can provide interesting alternatives for the identification of suitable candidates for various applications 17 or the determination of key structural and functional features, and it has been successfully employed in MOF research over the last years. [18][19][20][21][22][23][24][25] Another possible alternative is to use ML for systematization and validation of multiple reported synthesis methods for a particular material, as it was recently demonstrated for an archetypal Copper-trimesic acid MOF, HKUST-1. [26][27][28] Another example, is the use of linear models to investigate the formation of defects on UiO-66 MOF and their impact on its catalytic and adsorptive performance.…”

Inverse Design of ZIFs through Artificial Intelligence Methods

“…[2] Massive high-throughput MOF screenings in conjunction with artificial intelligence (AI) techniques, such as machine learning (ML), have proved to constitute a very powerful toolset that can extract complex correlations between the structure of a nanoporous solid family and its properties. [3], [4], [5], [6] However, even with a convenient MOF-performance correlation at our disposal, the design of materials for targeted separations will still be limited to a trial-and-error exploration of the materials space, albeit assisted by an improved chemical intuition. Therefore, the design of new materials calls for the inverse direction, which is the target-property -to-MOFstructure prediction.…”