Chemometrics in Biospectroscopy

Abstract

Vibrational spectroscopic techniques have been used for decades for predicting chemical composition of biological samples. For the establishment of prediction models, multivariate calibration techniques are often applied, since the use of single variables or combinations of few variables are not sufficient or robust enough. Multivariate techniques need to handle covariation among variables. We observe an especially high degree of covariation between variables in vibrational spectroscopy: Covariation is present (i) among neighbouring channels providing signals from the same spectral band, (ii) among spectral bands that are due to the same chemical bond, and (iii) finally among chemical bonds that arise from the same biomolecule. Consequently, the underlying dimensions in a given data set are rather few compared to the number of variables (channels) measured, also in cases where the number of samples is quite large. Multivariate methods based on latent variables that relate directly to underlying phenomena and mechanisms, are especially popular in vibrational spectroscopy. By using latent variables, the variations of objects can be studied with respect to these underlying phenomena and variation patterns can be studied in the light of chemical bonds and biomolecules. Another data modeling issue relates to the preprocessing of spectra: Vibrational spectra of biological materials are plagued by unwanted variability effects such as scatter effects in infrared spectroscopy or fluorescence effects in Raman spectroscopy. Calibration models often become simpler and even better when these effects are removed from the spectra before the calibration model is established. In addition, the interpretation of the results in terms of chemical bands is simpler and more reliable when the spectra are preprocessed. In this chapter, several issues related to data analysis of vibrational spectra of biological samples are discussed: (i) model‐based preprocessing techniques based on extended multiplicative signal correction (EMSC) are presented. Different variants of EMSC are discussed and a new extension of EMSC especially suited for the correction of fluorescence effects in Raman spectroscopy is introduced. (ii) Different aspects related to multivariate calibration and classification are discussed with a special focus on partial least squares regression (PLSR) and validation.

Quantitative determination of fatty acid compositions in micro-encapsulated fish-oil supplements using Fourier transform infrared (FTIR) spectroscopy

“…Prior to performing PCA, each spectrum was pre-processed by calculating 2nd derivatives using a 9-point Savitzky-Golay algorithm (Savitzky & Golay, 1964), and subsequently corrected by extended multiplicative scatter correction (EMSC) method (Kohler, Afseth, & Martens, 2010;Kohler, Kirschner, Oust, & Martens, 2005;Martens, Nielsen, & Engelsen, 2003) using only the spectral ranges that contain band information relevant to the lipid and protein components (i.e., 3100-2800 and 1800-950 cm À1 ). Outliers were removed based upon the samples with high residual variance across all principal components (PCs).…”

Spectral pre-processing for biomedical vibrational spectroscopy and microspectroscopic imaging

“…A second derivative spectrum is obtained from the sample spectrum. 3. At the precise wavenumber positions of water vapor lines, preferably in the spectral region between 1750-1900 cm -1 , second derivative intensities are extracted for both, the sample and water vapor spectrum.…”

“…Pre-processing has been shown to be of crucial importance for subsequent data mining tasks. In fact, it is now widely recognized that quantitative and classification models developed on the basis of pre-processed data generally perform better than models that solely use raw data [1][2][3]. With this review it is intended to explore the concepts and techniques of pre-processing methods and to discuss the applicability of distinct pre-processing techniques in the field of biomedical IR and Raman spectroscopy.…”

“…In vibrational spectroscopy this is useful to compensate for the differences in sample quantity or a different optical pathlength. In data mining tasks involving spectral distance measurements normalization has been shown to improve the accuracy and efficiency of the models [1][2][3]. 3.…”

“…In data mining tasks involving spectral distance measurements normalization has been shown to improve the accuracy and efficiency of the models [1][2][3]. 3. Filtering -in frequency analyses filtering is known as a group of methods that removes unwanted frequencies from the signals to be analyzed.…”