Multiresolution 3D-DenseNet for Chemical Shift Prediction in NMR Crystallography

We have developed a deep learning algorithm for chemical shift prediction for atoms in molecular crystals that utilizes an atom-centered Gaussian density model for the 3D data representation of a molecule. We define multiple channels that describe different spatial resolutions for each atom type that utilizes cropping, pooling, and concatenation to create a multiresolution 3D-DenseNet architecture (MR-3D-DenseNet). Because the training and testing time scale linearly with the number of samples, the MR-3D-DenseNet can exploit data augmentation that takes into account the property of rotational invariance of the chemical shifts, thereby also increasing the size of the training data set by an order of magnitude without additional cost. We obtain very good agreement for ¹³C, ¹⁵N, and ¹⁷O chemical shifts when compared to ab initio quantum chemistry methods, with the highest accuracy found for ¹H chemical shifts that is comparable to the error between the ab initio results and experimental measurements. Principal component analysis (PCA) is used to both understand these greatly improved predictions for ¹H , as well as indicating that chemical shift prediction for ¹³C, ¹⁵N, and ¹⁷O, which have far fewer training environments than the ¹H atom type, will improve once more unique training samples are made available to exploit the deep network architecture.

31P-dephased, 13C-detected REDOR for NMR crystallography at natural isotopic abundance

Typically, the process of NMR-based structure determination relies on accurately measuring a large number of internuclear distances to serve as restraints for simulated annealing calculations. In solids, the rotational-echo double-resonance (REDOR) experiment is a widely used approach to determine heteronuclear dipolar couplings corresponding to distances usually in the range of 1.5-8Å. A challenge in the interpretation of REDOR data is the degeneracy of symmetric subunits in an oligomer or equivalent molecules in a crystal lattice, which produce REDOR trajectories that depend explicitly on two or more distances instead of one. This degeneracy cannot be overcome by either spin dilution (for molecules containing 31P, 19F and other highly abundant nuclei) or selective pulses (in the case where there is chemical shift degeneracy). For small, crystalline molecules, such as phosphoserine, we demonstrate that as many as five inter-molecular distances must be considered to model 31P-dephased REDOR data accurately. We report excellent agreement between simulation and experiment once lattice couplings, 31P chemical shift anisotropy, and radio-frequency field inhomogeneity are all taken into account. We also discuss the systematic inaccuracies that may result from approximations that consider only the initial slope of the REDOR trajectory and/or that utilize a two- or three-spin system. Furthermore, we demonstrate the applicability of 31P-dephased REDOR for validation or refinement of candidate crystal structures and show that this approach is especially informative for NMR crystallography of 31P-containing molecules.

Solid state NMR studies of gels derived from low molecular mass gelators

Since its invention more than six decades ago, nuclear magnetic resonance (NMR) spectroscopy has evolved as an inevitable part of chemical as well as structural analysis of small molecules, polymers, biomaterials and hybrid materials. In the solution state, due to the increased viscosity of complex viscoelastic fluids such as gels, liquid crystals and other soft materials, the rate of molecular tumbling is reduced, which in turn affects the chemical shift anisotropy, dipolar and quadrupolar interactions. As a consequence the solution state NMR spectra show broad lines, and therefore, extracting detailed structural information is a challenging task. In this context, solid state (SS) NMR has the ability to distinguish between a minute amount of polymorphic forms, conformational changes, and the number of non-equivalent molecules in an asymmetric unit of a crystal lattice, and to provide both qualitative as well as quantitative analytical data with a short-range order. Therefore, SS NMR has continued to evolve as an indispensable tool for structural analysis and gave birth to a new field called NMR crystallography. Solid state cross polarization (CP) and high resolution (HR) magic angle spinning (MAS) NMR spectroscopy has been used to study weak interactions in polymer gels. However, the application of SS NMR spectroscopy to study gels derived from low molecular weight gelators has been limited until recently. In this review, we will focus on the importance of solid state NMR spectroscopy in understanding and elucidating the structure of supramolecular gels derived from low molecular weight gelators with selected examples.

NMR crystallography of ezetimibe co-crystals

An efficient, simplified protocol for solvent-drop assisted co-crystal preparation of ezetimibe (a drug for the treatment of primary hypercholesterolemia) with both imidazole and l-proline has been derived. The structures of the white powders were successfully solved via "NMR crystallography" combining solid-state NMR, powder X-ray diffraction and DFT chemical shift computations. Detailed insights into the likely crystallization mechanism were obtained from competition experiments, where efficient co-crystallization was feasible using ezetimibe monohydrate as precursor indicating that the crystal water acts as "molecular catalyst". It was also found that co-crystallization of imidazole is favored over l-proline, thus suggesting a clear preference of neutral hydrogen bonds compared to charge-assisted motifs.