Methods and applications of NMR crystallography for crystalline and amorphous pharmaceutical compounds

Atomic-level structure is essential for understanding chemical function, reactivity, and stability. Traditional techniques like single-crystal and powder X-ray diffraction are widely used for structure determination but often struggle with key limitations, particularly in resolving light atoms such as hydrogen and in analysing disordered or amorphous materials. While neutron diffraction and electron-based methods provide partial alternatives, they come with high sample and instrumentational demands and limited applicability. In contrast, solid-state nuclear magnetic resonance (NMR) offers a powerful alternative approach to atomic-level structure determination by probing local atomic environments and enabling structural insights regardless of crystallinity. Although many innovative applications of NMR crystallography have so far been reported in the literature in the context of organic, inorganic, and biological materials, the broader application of the method often remains hindered due to technical challenges. In this work, chemical shift based NMR crystallography was employed to determine atomic-level structures of organic molecular solids using advanced NMR techniques, fast magic angle spinning (MAS), and modern computational tools including machine learning. A primary focus was the development of 1H-detected strategies to accelerate data acquisition while maintaining high spectral resolution, which is crucial for efficient structure determination and validation. A major technical challenge addressed was the broadening of 1H resonances in solids, which limits spectral resolution and impedes assignment. By investigating the origins of proton linewidth under ultrafast MAS, it was found that inhomogeneous interactions (either anisotropic bulk magnetic susceptibility, ABMS, or structural disorder) and residual homonuclear dipolar couplings often dominate. Novel deep learning approaches, such as PIPNet and PIPNet2D, were developed to infer high-resolution 1H spectra free of residual homogeneous broadening, therefore significantly improving the 1H resolution in 1D and 2D datasets. These methodological advances enabled the high-resolution structural analysis and validation of crystalline molecular solids, including pharmaceutical compounds. Using a combination of fast MAS 1H-detected NMR, probabilistic assignment, and machine-learned chemical shifts, the crystal structures of these compounds could be validated more rapidly than traditional NMR crystallography workflows. Furthermore, NMR crystallography was applied to four highly relevant fully amorphous drug formulations, demonstrating that structural features such as hydrogen bonding and conformational preferences can be statistically characterized by matching experimental chemical shift distributions with predicted shifts from machine-learned models and molecular dynamics-generated structures. This approach provides valuable insights into stabilization mechanisms in disordered systems and expands th