Templated\nNucleation of Clotrimazole and Ketoprofen\non Polymer Substrates

The use of different template surfaces in crystallization\nexperiments\ncan directly influence the nucleation kinetics, crystal growth, and\nmorphology of active pharmaceutical ingredients (APIs). Consequently,\ntemplated nucleation is an attractive approach to enhance crystal\nnucleation kinetics and preferentially nucleate desired crystal polymorphs\nfor solid-form drug molecules, particularly large and flexible molecules\nthat are difficult to crystallize. Herein, we investigate the effect\nof polymer templates on the crystal nucleation of clotrimazole and\nketoprofen with both experiments and computational methods. Crystallization\nwas carried out in toluene solvent for both APIs with a template library\nconsisting of 12 different polymers. In complement to the experimental\nstudies, we developed a computational workflow based on molecular\ndynamics (MD) and derived descriptors from the simulations to score\nand rank API–polymer interactions. The descriptors were used\nto measure the energy of interaction (EOI), hydrogen bonding, and\nrugosity (surface roughness) similarity between the APIs and polymer\ntemplates. We used a variety of machine learning models (14 in total)\nalong with these descriptors to predict the crystallization outcome\nof the polymer templates. We found that simply rank-ordering the polymers\nby their API–polymer interaction energy descriptors yielded\n92% accuracy in predicting the experimental outcome for clotrimazole\nand ketoprofen. The most accurate machine learning model for both\nAPIs was found to be a random forest model. Using these models, we\nwere able to predict the crystallization outcomes for all polymers.\nAdditionally, we have performed a feature importance analysis using\nthe trained models and found that the most predictive features are\nthe energy descriptors. These results demonstrate that API–polymer\ninteraction energies are correlated with heterogeneous crystallization\noutcomes.