A quantitative computational framework for particle descriptors of pharmaceutical compounds

Beyond particle size, properties such as shape (morphology), surface chemistry, and roughness, have a considerable impact on powder behaviour, ultimately influencing the performance of pharmaceutical products. CCDC tools provide a way to calculate these particle properties for a given solid form by utilising only its crystal structure. This field of “particle informatics” currently lacks a large-scale reference describing what is “typical” for these computational descriptors across real-world, pharmaceutically relevant crystals. In this study, we present a computational workflow and a comprehensive particle property database derived from ~16,000 crystal structures from the Cambridge Structural Database (CSD) Drug Subset and three proprietary industrial datasets (AstraZeneca, GSK, and Pfizer). We predict particle morphologies and calculate facet- and particle-level descriptors using geometric methods and energetic (attachment energy) models with multiple force fields. Our analysis reveals systematic differences between public and proprietary datasets: the CSD Drug Subset is more diverse in multicomponent systems (salts, hydrates, co-crystals), resulting in higher surface polarity and interaction propensity, whereas proprietary datasets are dominated by single-component forms and exhibit rougher, more lipophilic surfaces. Increasing molecular complexity further drives roughness and chemical diversity, while morphology prediction biases highlight the sensitivity of shape classification to model choice. This work establishes a benchmark for the general application of “particle informatics” methods, providing a baseline of descriptor distributions and comparisons between public and proprietary datasets, to facilitate further advances in this growing field.