An Ecosystem for Digital Reticular Chemistry

Jablonka, Kevin Maik; Rosen, Andrew; Krishnapriyan, Aditi S.; Smit, Berend

doi:10.1021/acscentsci.2c01177

Cited by 15 publications

(20 citation statements)

References 121 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since RACs do not require the 3D Cartesian coordinates of atoms as input, they can be considered as a structure-agnostic descriptor. RACs of MOFs are obtained using the ( 60 ) package. XGBoost 61 model is used to make predictions using those handcrafted features.…”

Section: Methodsmentioning

confidence: 99%

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

et al. 2023

View full text Add to dashboard Cite

Metal−organic frameworks (MOFs) are materials with a high degree of porosity that can be used for many applications. However, the chemical space of MOFs is enormous due to the large variety of possible combinations of building blocks and topology. Discovering the optimal MOFs for specific applications requires an efficient and accurate search over countless potential candidates. Previous high-throughput screening methods using computational simulations like DFT can be time-consuming. Such methods also require the 3D atomic structures of MOFs, which adds one extra step when evaluating hypothetical MOFs. In this work, we propose a structure-agnostic deep learning method based on the Transformer model, named as MOFormer, for property predictions of MOFs. MOFormer takes a text string representation of MOF (MOFid) as input, thus circumventing the need of obtaining the 3D structure of a hypothetical MOF and accelerating the screening process. By comparing to other descriptors such as Stoichiometric-120 and revised autocorrelations, we demonstrate that MOFormer can achieve state-of-the-art structureagnostic prediction accuracy on all benchmarks. Furthermore, we introduce a self-supervised learning framework that pretrains the MOFormer via maximizing the cross-correlation between its structure-agnostic representations and structure-based representations of the crystal graph convolutional neural network (CGCNN) on >400k publicly available MOF data. Benchmarks show that pretraining improves the prediction accuracy of both models on various downstream prediction tasks. Furthermore, we revealed that MOFormer can be more data-efficient on quantum-chemical property prediction than structure-based CGCNN when training data is limited. Overall, MOFormer provides a novel perspective on efficient MOF property prediction using deep learning.

show abstract

Section: Methodsmentioning

confidence: 99%

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

et al. 2023

View full text Add to dashboard Cite

show abstract

“…As MOFs can be decomposed into their building blocks via recently developed tools like MOFid, [ 148 ] moffragmentor, [ 149 ] mBUD, [ 150 ] and MOFseek, [ 151 ] the MOF databases can be queried with the extracted building blocks to identify and probe MOFs with similar/identical building units of high‐performing MOF‐based membranes. These capabilities can be combined with ML models predicting MOFs with high thermal and solvent‐removal, [ 152 ] mechanical, [ 143 ] water stability [ 153 ] to identify materials that can potentially find use in industrial applications.…”

Section: Discussionmentioning

confidence: 99%

A New Era of Modeling MOF‐Based Membranes: Cooperation of Theory and Data Science

Demir,

Keskin

2023

Macro Materials & Eng

View full text Add to dashboard Cite

Membrane‐based separation can offer significant energy savings over conventional separation methods. Given their highly customizable and porous structures, metal–organic frameworks‐ (MOFs) are considered as next‐generation membrane materials that can bring about high separation performance and energy efficiency in various separation applications. Yet, the enormously large number of possible MOF structures necessitates the development and implementation of efficient modeling approaches to expedite the design, discovery, and selection of optimal MOF‐based membranes via directing the experimental efforts, time, and resources to the potentially useful membrane materials. With the recent developments in the field of atomic simulations and artificial intelligence methods, a new era of membrane modeling has started. This review focuses on the recent advances made and key strategies used in the modeling of MOF‐based membranes and highlight the huge potential of combining atomistic modeling of MOFs with machine learning to explore very large number of MOF membranes and MOF/polymer composite membranes for gas separation. Opportunities and challenges related to the implementation of data‐driven approaches to extract useful structure–property relations of MOF‐based membranes and to produce design principles for the high‐performing MOF‐based membranes are discussed.

show abstract

“…As recently conceptualized by Lyu et al, a synergistic workflow combining computational discovery and experimental validation can push material discovery to the next stage. 7,8 But to efficiently guide experimental discoveries, computational chemists are facing two major challenges: generating reliably more structures and evaluating them with fast and accurate models.…”

Section: Introductionmentioning

confidence: 99%

“…For the process to be viable, materials need to perform even better, and many studies focus on synthesizing ever more selective materials by leveraging all chemical intuitions around noble gas adsorption properties. − In order to speed the discovery process of novel materials with key properties, computational screening can identify factors explaining the performance and preselect candidates for further experimental studies. As recently conceptualized by Lyu et al, a synergistic workflow combining computational discovery and experimental validation can push material discovery to the next stage. , But to efficiently guide experimental discoveries, computational chemists are facing two major challenges: generating reliably more structures and evaluating them with fast and accurate models.…”

Section: Introductionmentioning

confidence: 99%

Enhancing Gas Separation Selectivity Prediction through Geometrical and Chemical Descriptors

Ren,

Coudert

2023

Chem. Mater.

View full text Add to dashboard Cite

Adsorption-based techniques for gas separation using nanoporous materials are widely used and hold a promising future, but systematic identification of the best-performing materials for a given application is still an open problem. For that task, we need to estimate selectivity at different operating conditions (temperature and pressure) on a large set of nanoporous structures. To this aim, we have developed a machine-learning-assisted screening process based on a fast grid calculation of interaction energies, in addition to newly designed geometrical descriptors to predict ambient-pressure selectivity. As a proof of concept, we tested our methodology for the separation of a 20:80 xenon/krypton mixture at 298 K and 1 atm in the nanoporous materials of the CoRE MOF 2019 database. Based on a train/test split of the data set, our model is promising with an RMSE of 2.5 on the ambient-pressure selectivity values of the test set and 0.06 on the log 10 of the selectivity. This method can thence be used to preselect the best performing materials for a more thorough investigation.

show abstract

An Ecosystem for Digital Reticular Chemistry

Cited by 15 publications

References 121 publications

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

MOFormer: Self-Supervised Transformer Model for Metal–Organic Framework Property Prediction

A New Era of Modeling MOF‐Based Membranes: Cooperation of Theory and Data Science

Enhancing Gas Separation Selectivity Prediction through Geometrical and Chemical Descriptors

Contact Info

Product

Resources

About