Adsorption Isotherm Predictions for Multiple Molecules in MOFs Using the Same Deep Learning Model
Adsorption Isotherm Predictions for Multiple Molecules in MOFs Using the Same Deep Learning Model
复制标题
DOI：
10.1021/acs.jctc.9b00940
复制
发表时间：
2020-02-01
影响因子：
5.5
通讯作者：
Gomez-Gualdron, Diego A.
中科院分区：
文献类型：
--
作者：
Anderson, Ryther;Biong, Achay;Gomez-Gualdron, Diego A.
调整金属有机框架（MOFs）的结构和化学性质，使其吸附性能的操纵，以适应特定的能源和环境应用。由于有数百万种可能的MOFs（已经合成了数万种），分子模拟经常用于快速评估大量MOFs的吸附性能。这使得随后的实验仅关注最有希望的MOF的一小部分。然而，在许多情况下，即使是分子模拟也变得非常耗时，这强调了在分子模拟工作之前需要替代筛选方法，例如机器学习。在这项研究中，作为概念证明，我们训练了一个神经网络-特别是多层wperceptron（MLP）-作为机器学习模型的第一个例子，该模型能够预测模型训练中不包括的不同分子的完整吸附等温线。为了实现这一点，我们训练我们的MLP上的“炼金术”的物种，只代表来自他们的力场参数的变量，预测负载的真实的吸附。用于训练的炼金术物种是小的、接近球形的、非极性的，使得能够预测与化学分离相关的类似的真实的分子，如氩、氪、氙、甲烷、乙烷和氮。MOFs也由简单的描述符表示（例如，几何性质和化学部分）。训练的模型表明，这六个吸附物在假设和现有的MOF的准确的吸附预测。这里提出的MLP预计不会“原样”应用于更复杂的吸附物，其性质在训练过程中未被考虑。然而，我们的研究结果说明了一种新的培训理念，可以建立在预测吸附等温线的目标，不仅在MOFs的数据库，而且在一系列相关的操作条件下的吸附物的数据库。
Tailoring the structure and chemistry of metal-organic frameworks (MOFs) enables the manipulation of their adsorption properties to suit specific energy and environmental applications. As there are millions of possible MOFs (with tens of thousands already synthesized), molecular simulation has frequently been used to rapidly evaluate the adsorption performance of a large set of MOFs. This allows subsequent experiments to focus only on a small subset of the most promising MOFs. In many instances, however, even molecular simulation becomes prohibitively time-consuming, underscoring the need for alternative screening methods, such as machine learning, to precede molecular simulation efforts. In this study, as a proof of concept, we trained a neural network-specifically, a multilayer wperceptron (MLP)-as the first example of a machine learning model capable of predicting full adsorption isotherms of different molecules not included in the training of the model. To achieve this, we trained our MLP on "alchemical" species, represented only by variables derived from their force-field parameters, to predict the loadings of real adsorbates. Alchemical species used for training were small, near-spherical, and nonpolar, enabling the prediction of analogous real molecules relevant for chemical separations such as argon, krypton, xenon, methane, ethane, and nitrogen. MOFs were also represented by simple descriptors (e.g., geometric properties and chemical moieties). The trained model was shown to make accurate adsorption predictions for these six adsorbates in both hypothetical and existing MOFs. The MLP presented here is not expected to be applied "as is" to more complex adsorbates with properties not considered during its training. However, our results illustrate a new philosophy of training that can be built upon with the goal of predicting adsorption isotherms in not only a database of MOFs but also a database of adsorbates and over a range of relevant operating conditions.