Study of the feasibility of using artificial neural networks in the search for new cocrystals

The development of new solid forms of bioactive agents, such as salts or cocrystals, has proven to be an important strategy for improving their physicochemical properties. Specifically, cocrystals allow the modification of such bioactive properties without altering the chemical structure of the compounds, as they are stabilized by noncovalent interactions. However, the development of this class of materials still requires extensive studies and experimental testing, including the selection of coformer molecules which, together with the bioactive agent, lead to cocrystal formation. In this context, artificial neural networks (ANN) have emerged as powerful tools in several areas, reducing costs and development time for this type of material. Thus, this work presents the preliminary results of using artificial neural networks in the prospection for new cocrystals. The results show that the use of ANN in such an application is feasible, with the developed model achieving an accuracy of 95.16%, precision of 97.37%, an F1-score of 96.1%, and an area under the ROC curve (ROC AUC) of 95.26% for a test set using 25 neurons in the hidden layer.