Molecular fingerprint and machine learning to accelerate design of high‐performance homochiral metal–organic frameworks

Abstract

Computational screening was employed to calculate the enantioseparation capabilities of 45 functionalized homochiral metal–organic frameworks (FHMOFs), and machine learning (ML) and molecular fingerprint (MF) techniques were used to find new FHMOFs with high performance. With increasing temperature, the enantioselectivities for (R,S)‐1,3‐dimethyl‐1,2‐propadiene are improved. The “glove effect” in the chiral pockets was proposed to explain the correlations between the steric effect of functional groups and performance of FHMOFs. Moreover, the neighborhood component analysis and RDKit/MACCS MFs show the highest predictive effect on enantioselectivities among the four ML classification algorithms with nine MFs that were tested. Based on the importance of MF, 85 new FHMOFs were designed, and a newly designed FHMOF, NO2‐NHOH‐FHMOF, with high similarity to the optimal MFs achieved improved chiral separation performance, with enantioselectivities of 85%. The design principles and new chiral pockets obtained by ML and MFs could facilitate the development of new materials for chiral separation.

Automatic Machine Learning Combined with High-Throughput Computational Screening of Hydrophobic Metal–Organic Frameworks for Capture of Methanol and Ethanol from the Air

“…The data was randomly split into two groups: the training set and the testing set, which make up 70% and 30% of the total data, respectively, using a random number generator. This proportion has been proved to be reliable by the works of Qiao and Li . The corresponding machine learning model was trained using the learning of the training set and then tested on the test set.…”

“…This proportion has been proved to be reliable by the works of Qiao 52 and Li. 53 The corresponding machine learning model was trained using the learning of the training set and then tested on the test set. Each ML method repeatedly predicted each performance five times, and the average Pearson correlation coefficient R 2 served as the final evaluation index of the model.…”

Minimal crystallographic descriptors of sorption properties in hypothetical MOFs and role in sequential learning optimization

“…Such computational tools allow us to identify significant correlations between nanoscale features and observable macroscale properties 14,15 , and to select the most suitable crystal for a given application case. A few representative examples are provided by gas-gas separation (D 2 /H 2 16 , O 2 /N 2 17 , CO/N 2 18 , CO 2 /H 2 19 , ethane/ethylene 20 , and other gas mixtures 21 ), the enantioselectivity of chemical compounds 22 , gas adsorption (CO 2 23 , CH 4 24 , H 2 25 , thiol 26 , organosulfurs 27 , and acetylene 28 ), and combinations thereof 29,30 . Several computational explorations of MOFs datasets have been carried out also for biomedical (drug delivery 31 ), mechanical (CO 2 Brayton cycle 32 and osmotic heat engine 33 ), and energy applications (heat pumps/chillers 34,35 and thermal energy storage 36 ).…”

Large-Scale Screening and Machine Learning for Metal–Organic Framework Membranes to Capture CO2 from Flue Gas

“…In the last twenty years, new materials metal–organic frameworks (MOFs), self-assembled by a wide range of organic links and metal nodes, have been considered to have potential for use in domains such as drug delivery [ 3 ], catalysis [ 4 ], gas storage [ 5 , 6 , 7 ], gas adsorption and separation [ 8 , 9 , 10 , 11 , 12 ] due to their excellent characteristics such as large surface area and high porosity. Commonly, MOFs can be applied as adsorbates and membranes for the separation of gas mixtures.…”

“…Figure S4: The calculation schematic diagram for the accuracy, sensitive, and specificity; Figure S5: The accuracy, sensitive, and specificity comparison of seven algorithms fo r S perm (CO 2 /N 2 ) and S perm (CO 2 /O 2 ) ; Figure S6: The diagram of k -fold cross validation; Figure S7: KNN algorithm model; Figure S8: SVM algorithm model; Figure S9: DT algorithm model; Figure S10: RF algorithm model; Figure S11: GBDT algorithm model; Figure S12: The leaf-wise tree growth schematic diagram of LGBM algorithm model; Figure S13: XGBoost algorithm model; Figure S14: Confusion matrix from best model; Figure S15: Atomistic structures of top-performance MOFMs. Table S1: Lennard–Jones parameters of metal–organic frameworks (MOFs); Table S2: Lennard–Jones parameters and charges of adsorbates; Table S3: Optimal hyperparameters for SVM, KNN, DT, RF, and GBDT; Table S4: Optimal hyperparameters for LGBM and XGBoost; Table S5: Evaluation of seven ML algorithm for P ; Table S6: Evaluation of seven ML algorithm for S perm : Table S7: Evaluation of XGBoost for P CO 2 and P O 2 ; Table S8: Evaluation of XGBoost for P N 2 ; Table S9: Evaluation of XGBoost for S perm (CO 2 /O 2 ) and S perm (CO 2 /N 2 ) ; Table S10: Seven top-performance MOFMs for CO 2 /N 2 /O 2 separation [ 9 , 37 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 ].…”