Prediction of Photochromism of Salicylideneaniline Crystals Using a Data Mining Approach

Salicylideneanilines (SAs) are photochromic compounds that undergo enol–keto photoisomerization in the solid state. Research over the past 60 years has revealed empirically that SAs with steric and planar conformations tend to be photochromic and nonphotochromic, respectively. However, increasing counterexamples in the recent literature raise questions about the nature of the relationship between structure and photochromism in SA crystals and whether the photochromism of SA crystals is predictable. This study is the first to construct a data set on SA crystals and conduct a comprehensive analysis to investigate the relationship between molecular and crystal structures and photochromism. A data mining approach revealed that the dihedral angle is the most dominant structural parameter for photochromism, followed by the Hirshfeld surface volume. SAs with neutral bulky hydrocarbon groups, such as the tert-butyl group, tend to be photochromic because such SAs have steric conformation and a loosely packed structure. In contrast, SAs with fluorine, pyridine, and pyrazine are less likely to be photochromic due to their planar conformation and densely packed structures. The photochromism of the SA crystals in our data set was predicted with high accuracy (>85%) using machine learning. The results of this study provide a useful reference for designing SA crystals with desired photochromic properties.