Jan Gröls1,Bernardo Castro-Dominguez1
University of Bath1
Jan Gröls1,Bernardo Castro-Dominguez1
University of Bath1
The use of Artificial Intelligence (AI) for designing molecular solids opens a new realm of possibilities to produce the next generation of semiconductors, optoelectronics, pharmaceuticals, and other advanced materials. In this work, we implemented a digital design method to produce binary supramolecular pharmaceutical assemblies via mechanochemistry. While these structures – namely co-crystals and co-amorphous solids (AS) – were selected to produce drugs with superior solubility and thus pharmacological performance; an intensive study on mechanochemical synthesis generated insights into this innovative, environmentally friendly synthetic method.<br/>During mechanochemical processes, mechanical forces trigger chemical reactions and occur without the presence of solvents. The supramolecular transformation of materials via solventless mechanochemistry has gained tremendous interest as reactants are not limited by their solubility in the solvent. Although, this has led to the discovery of new materials and reaction pathways, our mechanistic understanding remains limited. Predicting and designing mechanochemical processes remains challenging as there is a vast number of potential combinations of reactants under unseen reacting conditions.<br/>To predict the supramolecular transformation of binary crystalline active pharmaceutical ingredients (APIs), we conducted over 1500 experiments in a high throughput platform to generate data systematically. Pairs of APIs and second crystalline components were strategically chosen and screened by solventless neat grinding in a Retsch MM 400 vibration mill. Powder X-ray diffraction and differential scanning calorimetry characterize the degree of crystallinity of all the samples and assess the formation of new crystal structures or the generation of an AS. These experimental results served as input, and together with more than 2000 molecular descriptors, a machine learning algorithm (XGBoost) was implemented to predict the tendency of two molecules to either co-crystallize or co-amorphize under mechanochemical conditions.<br/>We identified the most important chemical descriptors for co-crystallization and amorphization, respectively, giving insights into the mechanistic behaviour of mechanochemical transformations. The predictive model led to the experimental discovery of new supramolecular structures, demonstrating the potential of applied AI in materials science.