Machine Learning Assisted Synthesis of Metal–Organic Nanocapsules

Abstract

Supplementary Figures and Tables Figure S1. Scheme of dataset splitting through randomly training/test splitting and hyperparameter tuning via 5-fold GridSearchCV. This process was repeated 5 times through changing random seed and the average of evaluation metrics was reported.

Artificial Chemist: An Autonomous Quantum Dot Synthesis Bot

“…[16] Among the rapidly growing variety of ML applications including expedited and enhanced analyses of complex reaction data sets, [17][18][19] deep reinforcement learning algorithms have recently been demonstrated to outperform highly trained human experts when utilized to predict the outcome, optimize the yield, and plan synthesis routes of various reactions in both supervised and self-optimizing systems. Such MLaccelerated strategies have been effectively applied to a large number of reactions spanning organic synthesis planning and optimization [20][21][22][23][24] and the formation of advanced materials including single crystal perovskites, [25] metal organic frameworks and nanocapsules, [26,27] gold nanoclusters, [16] and lead sulfide QDs. [28] Capitalizing on the recent progress of ML-enhanced optimization algorithms, a smart QD manufacturing strategy relying on decision-making algorithms and NNs trained on experimentally-measured QD properties can significantly accelerate the synthetic path discovery, optimization, and continuous manufacturing of colloidal QDs with precision-tailored optoelectronic properties.…”

Failure-experiment-supported optimization of poorly reproducible synthetic conditions for novel lanthanide metal-organic frameworks with two-dimensional secondary building units

“…We next turned our attention to extracting the dominant factors responsible for the poor reproducibility using decision tree analysis, which is considered to be one of the most interpretable machine-learning techniques. 14,19,21 Initially, the experimental data and cluster analysis results were linked together in a text file, after which the data file was analyzed using the decision tree technique, where the objective variables were the crystal phases assigned by cluster analysis of the PXRD patterns, and the explanatory variables were the synthetic parameters (see the Decision Tree Analysis section in the SI). The results presented in Figure 4a suggest that the most suitable synthetic conditions for the preparation of KGF-3 are as follows: Ligand solution concentration, 18-22 mM; cooling time, >12 h; and metal salt source, company A.…”

“…[11][12][13] In particular, several attempts to use machine learning to search for the crystallization conditions of materials have been reported, and it is beginning to be regarded as a powerful method. [14][15][16][17][18][19][20][21] While the application of machine learning techniques to predict crystallization conditions for nanoporous materials seems promising, its application in the exploration of novel materials remains limited, and no studies have introduced machine learning as a tool for the preparation of unknown MOFs. This is partially due to a lack of training data; open databases for the exploration of unknown MOFs are limited and the generation of such training data is expensive.…”

Failure-experiment-supported optimization of poorly reproducible synthetic conditions for novel lanthanide metal-organic frameworks with two-dimensional secondary building units

“…We next turned our attention to extracting the dominant factors responsible for poor reproducibility by employing decision tree analysis, which is considered to be one of the most interpretable machine-learning techniques. 14,19,21 Initially, the experimental data and cluster analysis results were linked together in a text file, after which the data file was analyzed using the decision tree technique, whereby the objective variables were the crystal phases assigned by cluster analysis of the PXRD patterns, and the explanatory variables were the synthetic parameters (see the Decision Tree Analysis section in the SI). The results presented in Figure 4a suggest that the most suitable synthetic conditions for the preparation of KGF-3 are as follows: Ligand solution concentration, 18-22 mM; cooling time, >12 h; and metal salt source, company A.…”

“…[11][12][13] In particular, several attempts using machine Please do not adjust margins Please do not adjust margins learning to search for crystallization conditions of materials have been reported, and it is now beginning to be regarded as a powerful method. [14][15][16][17][18][19][20][21] While the application of machine learning techniques to predict crystallization conditions for nanoporous materials seems promising, its application in the exploration of novel materials remains limited, and no studies thus far have introduced machine learning as a tool for the preparation of unknown MOFs. This is partially due to a lack of training data; open databases for the exploration of unknown MOFs are limited and the generation of such training data is expensive.…”