Different
artificial neural network (ANN) models have been developed and examined
for prediction of cocrystal properties based on pure component physical
properties only. From the molecular weight, melting temperature, melting
enthalpy, and melting entropy of the pure compounds, the corresponding
melting properties of the cocrystals and the cocrystal ideal solubility
have been successfully predicted. Notably, no information whatsoever
about the cocrystals is needed, besides the identification of the
two compounds from which the cocrystal is formed. In total, 30 cocrystal
systems of 8 different model components, namely, theophylline, piracetam,
gabapentin-lactam, tegafur, nicotinamide, salicylic acid, syringic
acid, and 4,4′-bipyridine, with distinct coformers have been
chosen as the model systems for the construction of ANN models. In
all the cases, 70% of the data points have been used to train the
model, and the rest were used to test the capability of the model
(as a validation set) as selected through a random selection process.
The training process was stopped with overall r
2 values above 0.986. In particular, the models capture how
the coformer structure influences the targeted physical properties
of cocrystals.