Molecular screening for solid–solid phase transitions by machine learning

Abstract

The solid–solid phase transition in molecular crystals is generally found by chance empirically. In this study, we constructed a machine learning framework to screen molecules that will exhibit solid–solid phase transitions in their crystalline states, based on positive-unlabeled learning. We trained classification models using the positive dataset we constructed manually and the unlabeled data extracted from the Cambridge Structural Database. The best classifier works as a suggester, and 9 substances among the suggested 113 molecules were found to exhibit solid–solid phase transitions according to the literature and experiments. The finding probability of 8.0% is much higher than the probability of phase transition in the database, suggesting the effectiveness of molecular selection by this workflow. We also found that the molecular structure is weakly related to the transition temperature by regression analysis. The findings of this study are useful for designing functional molecular crystals with solid–solid phase transitions.