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

Abstract

The capture of low concentration alcohol VOCs (methanol and ethanol) from the air has also attracted more and more attention. In this work, high-throughput computational screening (HTCS) and machine learning (ML) methods based on molecular simulations were used to investigate the adsorption properties of methanol and ethanol in 31 399 hydrophobic metal–organic frameworks (MOFs). First, the structure–performance relationship of MOFs was successfully established through univariate analysis, and the key descriptors identified were LCD and Q 0 st. Five ML methods, Decision Tree (DT), Random Forest (RF), Back Propagation Neural Network (BPNN), Support Vector Machines (SVM), and Tree-based Pipeline Optimization Tool (TPOT), were used to predict the adsorption performance of MOFs. The automatic machine learning (Auto-ML) algorithm TPOT has the best prediction effect on the TSN of methanol and ethanol, with R 2 values of 0.852 and 0.945, respectively. The accuracy of the ML model was further improved using the random search method. Analysis of the algorithms has revealed that GBR and RFR have the highest prediction accuracy and frequency, respectively, for the MOF–methanol and MOF–ethanol systems. Ten MOF materials with excellent adsorption properties (0.002 mol/kg ≥ N CH3OH ≥ 0.001 mol/kg, 0.068 mol/kg ≥ N C2H5OH ≥ 0.016 mol/kg; 420.67 ≥ S CH3OH ≥ 214.29, 3.2 × 106 ≥ S C2H5OH ≥ 8.5 × 103) were selected successfully. After analysis of their adsorption sites, it was found that the primary adsorption sites for methanol and ethanol are located near the amino and halogen groups, and the different metal centers showed great influence on the adsorption capacity of MOFs for two kinds of alcohol molecules through the analysis of their structural commonness. This work can serve as a roadmap for experimental synthesis, innovative design of MOFs, and the development of new ML algorithms.

An auto‐configurable machine learning framework to optimize and predict catalysts for CO2 to light olefins process

“…The prediction accuracy of the AutoML‐DE model for various output parameters is evaluated by the root mean square error (RMSE) and determination coefficient ( R 2 ), which are defined as shown in Equations () and () 33 . R 2 is an indicator of the similarity between estimated and measured values, while RMSE serves as the standard deviation between predicted and true values and is widely used to measure the degree of difference between the two 20 where n is the number of samples; and represent the true and predicted values of the i th data; and denotes the average of the dataset.…”

“…Professor Teng Zhou's team 17–19 have employed this approach to analyze the operational efficacy of covalent organic frameworks in methane utilization, highlighting that AutoML can streamline the modeling process, enable automatic configuration of ML model parameters, and overcome the constraint of expert involvement. Similarly, Zhang et al 20 studied an AutoML algorithm to improve the prediction accuracy of the adsorption performance of metal‐organic frameworks. They found that the AutoML algorithm can not only enhance the reliability of prediction results, but also avoid overfitting in the model.…”

Predicting and Evaluating Different Pretreatment Methods on Methane Production from Sludge Anaerobic Digestion via Automated Machine Learning with Ensembled Semisupervised Learning

“…Since the range of data varies widely, the comparison will be inconsistent without normalization. Hence, this work used a row scale normalization from 1 to 10 approach in data processing, which has also been used in previous work . The unified units facilitated meaningful comparisons and analyses, ensuring the data set’s quality and integrity.…”

“…The AutoML can automatically select the suitable model to predict methane production and determine the related hyperparameters without artificial intervention. Meanwhile, AutoML can improve the reliability of prediction results by avoiding overfitting and other issues . However, the limited availability of data often restricts the application of data-driven models in anaerobic bioprocess research, especially in the context of novel bioprocess testing and initial experimental endeavors …”