Relationships between Changes in Guest Ion Properties and in the Host Framework Topology in Ionic Coordination Polymers

In order to identify possible general relationships between changes in the framework topology and changes in the extra framework ion properties, a set consisting of 2202 crystal structures of ionic coordination polymers was extracted from the Cambridge Structural Database. Changes in ion properties served as independent variables for several machine learning models trained to predict the changes in framework dimensionality, topological density, and average ring size of the framework’s tiling. The trained classifiers showed acceptable predictive performance with F1 score in the range 0.4 ÷ 0.6 and were subjected to the validation tests, which confirmed that they fit the data significantly better than by chance. Subsequent feature importance analysis of the classifiers revealed a set of the ion properties being important for prediction of the changes in corresponding framework characteristics in the extracted set of crystal structures. It is shown that in general changes in molecular surface area and molecular flexibility of the guest ions are essential for predicting changes in selected topological characteristics of a framework. Case studies were conducted for several sets of crystal structures with frameworks that are observed to host significant variety of counter ions. The decision tree classifiers allowed us to discover the ion properties determining topological characteristics in particular frameworks.

为探究骨架拓扑结构（framework topology）变化与骨架外离子性质变化间可能存在的普适性关系，我们从剑桥晶体结构数据库（Cambridge Structural Database）中提取得到一套包含2202个离子配位聚合物（ionic coordination polymers）晶体结构的数据集。以离子性质变化作为自变量，训练多款机器学习模型（machine learning models）以预测骨架维度、拓扑密度以及骨架平铺平均环尺寸的变化。所训练得到的分类器表现出可接受的预测性能，其F1分数（F1 score）区间为0.4至0.6，并通过了验证测试。测试结果证实，相较于随机猜测，这些模型对数据的拟合效果显著更优。后续针对该批分类器的特征重要性分析显示，在本次提取的晶体结构数据集内，存在一组离子性质，其对预测对应骨架特征的变化具有重要意义。研究表明，总体而言，客体离子的分子表面积变化与分子柔性变化，是预测骨架选定拓扑特征变化的核心要素。针对多组包含可容纳多种抗衡离子骨架的晶体结构数据集，我们开展了案例研究。借助决策树分类器（decision tree classifiers），我们得以发掘出决定特定骨架拓扑特征的关键离子性质。