Document Type

Theses, Ph.D


Available under a Creative Commons Attribution Non-Commercial Share Alike 4.0 International Licence



Publication Details

A thesis successfully submitted for the Degree of Doctor of Philosophy.


Raman spectroscopy is a growing technology in the fields of in-vitro drug and nanoparticle screening. The label free capability provided by vibrational spectroscopy, as well as the ability of the technique to probe the chemical nature of samples, makes it a good candidate for use in these fields. Crucial to the progress of these methods is the development and validation of robust and accurate multivariate statistical analysis protocols. In this thesis, both established and novel methods are examined using both real and simulated datasets. In particular, simulated datasets are used to validate and assess the accuracy of these methods in a spectroscopic setting. Firstly, partial least squares regression (PLSR) is examined using a simulated model based on real experimental data. This is applied to investigate the application of the algorithm to continuously varying data with known spectral perturbations introduced over a range of concentrations and responses. The results show that, while PLSR is valid for some dose ranges, sub-lethal, low concentrations and thus subtle spectral changes in the data may lead to difficulties in model construction. Multiple trends present in the data were also investigated and possible model error based on spectral bleedthrough in the regression coefficients RCs is explored. Principal component analysis (PCA) was also investigated using simulated datasets based on known changes in the data. Some of the limitations of PCA for data partitioning and trend analysis are overcome by a novel variant termed, ‘seeded’ PCA. 1st and 2nd derivative data is also explored for improvements in Raman spectral analysis using seeded PCA.


https:/ /