Multiway NMR mixture analysis

It is hard to overstate the importance of mixture analysis: it underpins progress in much of chemistry, biology and pharmacy. Paradoxically, the best method we have for obtaining information on chemical structure, NMR, struggles with all but the simplest mixtures. NMR spectra are relatively easily interpreted by experienced chemists, but only for pure compounds. However, NMR has great potential for efficient mixture analysis; it is both non-destructive and non-invasive, allowing the study of completely intact mixtures. The limited success enjoyed by NMR in mixture analysis to date stems not so much from its inherent limitations, as from failure to integrate experiment and data analysis optimally. The principal reason for the difficulty of mixture analysis is one that we are all familiar with: information overload. A single NMR spectrum contains a wealth of information, accessible through e.g. chemical shifts, multiplet structure and NOEs. The very richness of NMR is its central problem: once we have multiple spectra superimposed, it rapidly becomes impossible to identify and assign individual signals. One answer is to harness the power of multivariate data analysis, and in particular multiway methods. The latter are very under-exploited: most multivariate analysis is two-way rather than multiway. Two-way methods such as PCA and ICA are very important, but have a crucial limitation. Used for unconstrained factorisation of an experimental dataset (for example a series of diffusion-weighted spectra of a mixture), even with common strong constraints such as non-negativity they will decompose the data into a set of distinct component spectra but those spectra will be arbitrary mixtures of the spectra of the mixture components, not the spectra themselves. If, however, we extend the dimensionality of the data, by adding a further source of variation, we can use multiway methods to extract the true, unmixed, component spectra, without the need for strong constraints such as statistical independence, thanks to the inherent uniqueness of multiway models. The flexibility of NMR allows us to develop pulse sequences to generate data that are specifically tailored for optimum multiway analysis.
In this studentship we will design a new family of NMR experiments that produce multilinear data tailored for the use of multiway algorithms, to extract spectra and other data for individual species and/or spin systems from intact mixtures. This has the potential to revolutionize the way we do mixture analysis by NMR.