The reverse process of calculating a protein structure from chemical shifts is to predict the chemical shifts from a structure. Shift prediction is in fact an essential part of chemical shift based structure calculation protocols. It allows an interactive cycle to be used in which the predicted shifts of the currently calculated structures are compared to the input shifts and the best structures can then be selected as starting structures for the next round of calculations. Of course, accurate chemical shift prediction would also be a useful route to chemical shift assignment in situations where the structure of a protein is known, but a researcher wants to use NMR to study the dynamics or intermolecular interactions of a protein.
The most accurate predictions of chemical shifts are achieved using quantum chemistry methods. However, this is a computationally very intensive approach which is currently only applied to small molecules or molecular fragments and not to whole proteins. Predictions of protein chemical shifts are mainly based on databases of high-resolution protein structures whose chemical shifts are known. A combination of local sequence and/or structural comparison to the query structure and machine learning methods are used to predict chemical shifts. Several programs which predict protein chemical shifts are listed on the software page.
In general, predictions are best for Cα and Cβ and worst for NH atoms. This means that the accurate prediction of a 1H-15N HSQC spectrum is still one of the most difficult tasks.