Molecular screening for solid–solid phase transition by machine learning

Solid–solid phase transition is a mechanism manifesting and regulating functionalities in molecular crystals. The phase transitions are generally found by chance empirically, and screening molecules theoretically to predict their existence is challenging. In this study, we screened for the possibility of solid–solid phase transitions by positive-unlabeled learning using molecular descriptors and verified the inference by finding new substances exhibiting solid–solid phase transitions in the solid state. We also found that the molecular structure of a substance is weakly related to the transition temperature but not the transition enthalpy. The findings of this study are useful for designing functional molecular crystals.