Machine Learning Prediction of Flavonoid Cocrystal Formation Combined with Experimental Validation

This study established a flavonoid cocrystal database, and four machine learning models [support vector machine (SVM), random forest (RF), logistic regression (LR), and artificial neural network (ANN)] were established to screen flavonoid cocrystal coformers based on three screening methods of the molecular descriptors [original, principal component analysis (PCA)-selected, and quantitative structure-property relationship (QSPR)-selected descriptors]. In addition, the apigenin-4,4'-bipyridine cocrystal was prepared and characterized based on the prediction of the models. At the same time, through the performance comparison between the models and the logistic model analysis of molecular descriptors related to the hydrogen bonds, it is concluded that the molecular descriptors based on the molecular structure of the flavonoid cocrystals (phenolic groups and nitrogen atoms, etc.) have a great influence on the models, such as nC, nN, nO, etc., and the formation of flavonoid cocrystal is closely related to the hydrogen bond. The excellent performance of the machine learning model in the flavonoid cocrystal database further confirms that it is a more scientific and reliable method to establish a single API cocrystal database and to develop the corresponding machine learning model.

Automated summary

This study established a flavonoid cocrystal database and successfully screened flavonoid cocrystal coformers using machine learning models and three screening methods. The results demonstrated the close relationship between the formation of flavonoid cocrystals and molecular descriptors such as phenolic groups and nitrogen atoms.