Statistical strategies to reveal potential vibrational markers forin vivoanalysis by confocal Raman spectroscopy

Abstract

The analysis of biological systems by spectroscopic techniques involves the evaluation of hundreds to thousands of variables. Hence, different statistical approaches are used to elucidate regions that discriminate classes of samples and to propose new vibrational markers for explaining various phenomena like disease monitoring, mechanisms of action of drugs, food, and so on. However, the technical statistics are not always widely discussed in applied sciences. In this context, this work presents a detailed discussion including the various steps necessary for proper statistical analysis. It includes univariate parametric and nonparametric tests, as well as multivariate unsupervised and supervised approaches. The main objective of this study is to promote proper understanding of the application of various statistical tools in these spectroscopic methods used for the analysis of biological samples. The discussion of these methods is performed on a set of in vivo confocal Raman spectra of human skin analysis that aims to identify skin aging markers. In the Appendix, a complete routine of data analysis is executed in a free software that can be used by the scientific community involved in these studies.

Review of Laser Raman Spectroscopy for Surgical Breast Cancer Detection: Stochastic Backpropagation Neural Networks

“…A wide variety of techniques have been investigated including wavelets [26], FFTs [27], first and second derivatives [28], and polynomial fitting [29]. At present the most commonly implemented baseline correction techniques involve some form of polynomial fitting [30]. An easily implemented method uses an asymmetrically reweighted penalized least squares smoothing algorithm (arPLS) developed by Baek et al [31].…”