Sustainable Co-Crystal Formation: AI-Driven Design and Experimental Validation

Sustainable Co-Crystal Formation: AI-Driven Design and Experimental Validation This thesis explores various aspects of co-crystallization in relation to the growing need for more sustainable processes in pharmaceutical formulation discovery. It begins by discussing the development of predictive tools designed to reduce the number of experimental tests, proceeds with the use of mechanochemistry and green chemistry approaches for synthesizing co-crystals, and concludes with the introduction of a novel and alternative platform for permeability evaluations, intended to replace animal testing in early-stage drug development. A primary focus is placed on understanding how chemical and geometrical factors guide molecular assembly in the solid state, an essential aspect of crystal engineering. However, traditional approaches often face challenges when dealing with complex pharmaceutical molecules, where stereocenters and diverse functional groups complicate the prediction of weak interactions. To address these challenges, alternative predictive approaches are proposed, leveraging machine learning models trained on tailored “co-crystallization descriptors” and chemical language processing techniques. In particular, the developed model DeepCocrystal introduces significant advancements in the application of AI for co-crystal design. By utilizing chemical language representations (SMILES) and uncertainty estimation, it enhances the accuracy of co-crystallization predictions. Rigorous validation of DeepCocrystal showed a balanced accuracy of 78% in realistic scenarios, outperforming existing models. The success of DeepCocrystal, verified through experimental synthesis, demonstrates its potential to streamline co-crystal discovery while reducing waste and costs. Additionally, experimental case studies (such as those involving tuberculosis co-crystals) illustrate the impact of co-crystallization on drug solubility and permeability. These results, however, emphasize the importance of accurately predicting these properties prior to synthesis. The novel Meso-fluidic Chip for Permeability Assessment enhances sustainable drug testing by offering an in vitro alternative to animal studies for permeation testing. Ultimately, this work highlights innovative predictive and experimental methods that contribute to a greener pharmaceutical development process. By integrating experimental techniques with AI-driven strategies, the thesis aims to transform co-crystallization design into a sustainable and efficient process, while advancing the broader field of supramolecular material discovery in alignment with eco-friendly practices.