Multivariate statistical methodologies applied in biomedical Raman spectroscopy: assessing the validity of partial least squares regression using simulated model datasets.

Raman spectroscopy is fast becoming a valuable analytical tool in a number of biomedical scenarios, most notably disease diagnostics. Importantly, the technique has also shown increasing promise in the assessment of drug interactions on cellular and subcellular levels, particularly when coupled with multivariate statistical analysis. However, with respect to both Raman spectroscopy and the associated statistical methodologies, an important consideration is the accuracy of these techniques and more specifically, the sensitivities which can be achieved, and ultimately the limits of detection of the various methods. The purpose of this study is thus the construction of a model simulated dataset with the aim of testing the accuracy and sensitivity of the partial least squares regression (PLSR) approach to spectral analysis. The basis of the dataset is the experimental spectral profiles of a previously reported Raman spectroscopic analysis of the interaction of the cancer chemotherapeutic agent cisplatin in an adenocarcinomic human alveolar basal epithelial cell-line, in vitro, and is thus reflective of actual experimental data. The simulated spectroscopic data are constructed by adding known perturbations which are independently linear in drug doses as well as cytological responses experimentally determined by a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) cytotoxicity assay. It is demonstrated that, through appropriate choice of dose range, PLSR against the respective targets can differentiate between the spectroscopic signatures of the direct chemical effect of the drug dose and the indirect cytological effect it produces.