Multivariate Analysis of NMR Spectroscopy: A Comparison Between MVA-Plots and SIMCA 18

Abstract

Many diagnoses for diseases rely on testing which can be invasive, expensive, time-consuming, or inaccurate 1,2. Metabonomic analyses of NMR spectra may be helpful methods of disease diagnosis3. This sort of analysis is made possible because the NMR spectroscopic datasets have great similarity, making them prime candidates for Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) and Principal Component Analysis (PCA)4. These allow us to identify trends in the data in a computationally efficient way by breaking down data sets into their principal components4. Diffusion Ordered Spectroscopy (DOSY) is a method of H-NMR spectroscopy which defines diffusion coefficients for the different resonances found within a mixture5. In this thesis, I perform DOSY NMR for three different gradient ramps, a linear ramp, a quadratic ramp and an exponential ramp. The linear ramp is the control, and takes measurements with the current increasing linearly. The quadratic ramp takes measurements with the current increasing quadratically. The exponential ramp takes measurements with the current increasing exponentially. The last spectrum is a high-powered diffusion probe. After performing DOSY NMR, I download and set up the MVA-plots code so that it can be compared to SIMCA 18 for PCA and OPLS. The linear ramp is found to have a good resolution for small and large molecules. The quadratic ramp appears to have the best all-around resolution for small and large molecules. The exponential ramp has high resolution for small molecules. The PCA and OPLS functions for MVA-plots and SIMCA 18 are found to produce highly similar graphs and have the same R2X[1] and R2X[2] values for PCA and the same R2X(cum), R2Y(cum) and Q2(cum) values for OPLS. These values being the same suggest that MVA-plots and SIMCA 18 have the same functionality for these functions.