Bailey Tom, Jackson Adam, Berbece Razvan-Antonio, Wu Kejun, Hondow Nicole, Martin Elaine
School of Chemical and Process Engineering, University of Leeds, Leeds LS2 9JT, U.K.
Zhejiang Provincial Key Laboratory of Advanced Chemical Engineering Manufacture Technology, College of Chemical and Biological Engineering, Zhejiang University, Hangzhou 310027, China.
J Chem Inf Model. 2023 Aug 14;63(15):4545-4551. doi: 10.1021/acs.jcim.3c00135. Epub 2023 Jul 18.
Predictive screening of metal-organic framework (MOF) materials for their gas uptake properties has been previously limited by using data from a range of simulated sources, meaning the final predictions are dependent on the performance of these original models. In this work, experimental gas uptake data has been used to create a Gradient Boosted Tree model for the prediction of H, CH, and CO uptake over a range of temperatures and pressures in MOF materials. The descriptors used in this database were obtained from the literature, with no computational modeling needed. This model was repeated 10 times, showing an average of 0.86 and a mean absolute error (MAE) of ±2.88 wt % across the runs. This model will provide gas uptake predictions for a range of gases, temperatures, and pressures as a one-stop solution, with the data provided being based on previous experimental observations in the literature, rather than simulations, which may differ from their real-world results. The objective of this work is to create a machine learning model for the inference of gas uptake in MOFs. The basis of model development is experimental as opposed to simulated data to realize its applications by practitioners. The real-world nature of this research materializes in a focus on the application of algorithms as opposed to the detailed assessment of the algorithms.
先前，针对金属-有机骨架（MOF）材料的气体吸收性能的预测筛选受到了来自各种模拟源的数据的限制，这意味着最终的预测结果取决于这些原始模型的性能。在这项工作中，我们使用实验气体吸收数据创建了一个梯度提升树模型，用于预测 MOF 材料在一系列温度和压力下对 H、CH 和 CO 的吸收。该数据库中使用的描述符是从文献中获得的，无需进行计算建模。该模型重复了 10 次，在运行过程中平均达到了 0.86，平均绝对误差（MAE）为±2.88wt%。该模型将提供一系列气体、温度和压力下的气体吸收预测，作为一站式解决方案，提供的数据基于文献中的先前实验观察结果，而不是模拟结果，模拟结果可能与实际结果不同。这项工作的目的是创建一个机器学习模型，用于推断 MOF 中的气体吸收。模型开发的基础是实验数据而不是模拟数据，以实现其在从业者中的应用。该研究材料的实际应用体现在对算法的应用关注，而不是对算法的详细评估。
