Tutorial: multivariate classification for vibrational spectroscopy in biological samples

Abstract

The use of vibrational spectroscopy techniques, such as Fourier-transform infrared (FTIR) and Raman spectroscopy, has been a successful method to study the interaction of light with biological materials and facilitate novel cell biology analysis. Disease screening and diagnosis, microbiological studies, forensic and environmental investigations make use of spectrochemical analysis very attractive due to its low cost, minimal sample preparation, non-destructive nature and substantially accurate results. However, there is now an urgent need for multivariate classification protocols allowing one to analyse biological-derived spectrochemical data in order to obtain accurate and reliable results. This is stimulated by the fact that applications of deep-learning algorithms of complex datasets are being increasingly recognized as critical towards extracting important information and visualizing it in a readily interpretable form. Hereby, we have constructed a protocol for multivariate classification analysis of vibrational spectroscopy data [FTIR, Raman and near-infrared (NIR)] highlighting a series of critical steps, such as pre-processing, data selection, feature extraction, classification and model validation. This is an essential aspect towards the construction of a practical spectrochemical analysis model for biological analysis in real-world applications, where fast, accurate and reliable classification models are fundamental.

ATR-FTIR spectroscopy in blood plasma combined with multivariate analysis to detect HIV infection in pregnant women

“…SVM is a binary linear classifier using a nonlinear step that transforms the input sample space into a feature space using a kernel function that maximizes the margins of separation between the sample groups 31 . In this study, a radial basis function (RBF) kernel was used since it is able to adapt to different data distributions 32 . The SVM classification rule is obtained by the following Eq.…”

Validation of biosignatures confirms the informative nature of fossil organic Raman spectra

“…All Raman spectra include spectral noise which commonly affects the shape of the baseline (see 33 for fossil examples). The application of a common adaptive baseline to different kinds of analyzed samples might affect some normalized spectra more than others: However, any impact this might have on interpretations is circumvented by evaluating large spectral data sets statistically (as in 3,4,6,[10][11][12]15) using ChemoSpace Principal Component or Discriminant Analyses (16,17,20,24). The conclusions of the papers (3,4,6,11,12,15) that Alleon et al (21) disputed were the result of such ordination methods applied to spectral intensities obtained at previously identified, informative band positions.…”

“…The routine used in the disputed papers corresponds to the multivariate analysis of Raman spectra in whole-tissue cancer diagnostics (16,(18)(19)(20), and the investigation of whole-tissue archeological remains (33). Such a statistical approach is a more robust route to generalized inferences than relying on qualitative analyses of small numbers of samples (16)(17)(18)(19)(20).…”

“…Raman scattering can be induced by excitation wavelengths ranging from the ultraviolet, to visible (such as 532 nm) and the near-infrared range, and all of these approaches produce reliable insights into patterns in the composition of biomaterials (quantitatively compared in 16). Raman spectra obtained under identical conditions for a given sample set can be standard-baselined, normalized, and analyzed by means of multivariate statistics (16)(17)(18)(19)(20). Ordination methods are most commonly used for comparative spectral analyses in the search for patterns in the total organic composition of complex biological samples (16)(17)(18)(19)(20).…”

“…Raman spectra obtained under identical conditions for a given sample set can be standard-baselined, normalized, and analyzed by means of multivariate statistics (16)(17)(18)(19)(20). Ordination methods are most commonly used for comparative spectral analyses in the search for patterns in the total organic composition of complex biological samples (16)(17)(18)(19)(20). A number of advantages, including time-efficient nondestructive analysis, wide availability and low operating cost of equipment, and utility of the results, make Raman and complementary types of light spectroscopy ideal for initial molecular assessments of paleontological and geological materials (3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15).…”