NMR crystallography for validating organic crystal structures

密度泛函理论代码的发展意味着现在计算给定晶体结构的核磁共振参数(如化学位移)相对简单。在最近研究剑桥结构数据库中的重复结构测定的工作中，我们发现许多情况下需要额外的实验证据，例如来自核磁共振的证据，以区分来自衍射研究的替代解决方案。鉴于核磁共振核的多样性，核磁共振可观测到的多样性(例如各向同性化学位移、四极耦合等)，以及可用的一维和二维核磁共振实验的多样性，这并不明显，然而，核磁共振实验最有效地区分正确和不正确的结构。将评估一些验证晶体结构的方法，从简单测量实验位移和计算位移之间的偏差到更复杂的方法，如贝叶斯推理，以定量确定不同数据集提供的信息。我们还将利用当地在机器学习方法方面的专业知识来识别结果中的模式，因为不同的情况可能需要不同的实验输入。

Developments in Density Functional Theory codes means that it is now relatively straightforward to calculate Nuclear Magnetic Resonance (NMR) parameters, such as chemical shifts, for a given crystal structure. In recent work examining repeat structure determinations in the Cambridge Structural Database, we have found many cases where additional experimental evidence, e.g. from NMR, is needed to distinguish between alternative solutions derived from diffraction studies. Given the variety of NMR nuclei, the variety of NMR observables (e.g. isotropic chemical shifts, quadrupolar couplings etc.), and the variety of 1D and 2D NMR experiments available it is not obvious, however, which NMR experiments distinguish most effectively between correct and incorrect structures. A number of approaches to validating crystal structures will be evaluated, ranging from simple measures of deviations between experimental and calculated shifts to more sophisticated methods e.g. Bayesian inference to determine quantitatively the information provided by different data sets. We will also exploit local expertise in machine learning approaches to identify patterns in results, since it is likely that different situations will require different experimental inputs.