Review of multidimensional data processing approaches for Raman and infrared spectroscopy

Abstract

Raman and Infrared (IR) spectroscopies provide information about the structure, functional groups and environment of the molecules in the sample. In combination with a microscope, these techniques can also be used to study molecular distributions in heterogeneous samples. Over the past few decades Raman and IR microspectroscopy based techniques have been extensively used to understand fundamental biology and responses of living systems under diverse physiological and pathological conditions. The spectra from biological systems are complex and diverse, owing to their heterogeneous nature consisting of bio-molecules such as proteins, lipids, nucleic acids, carbohydrates etc. Sometimes minor differences may contain critical information. Therefore, interpretation of the results obtained from Raman and IR spectroscopy is difficult and to overcome these intricacies and for deeper insight we need to employ various data mining methods. These methods must be suitable for handling large multidimensional data sets and for exploring the complete spectral information simultaneously. The effective implementation of these multivariate data analysis methods requires the pretreatment of data. The preprocessing of raw data helps in the elimination of noise (unwanted signals) and the enhancement of discriminating features. This review provides an outline of the state-of-the-art data processing tools for multivariate analysis and the various preprocessing methods that are widely used in Raman and IR spectroscopy including imaging for better qualitative and quantitative analysis of biological samples.

Detection and classification of Bacteria using Raman Spectroscopy Combined with Multivariate Analysis

“…Additionally, linear discriminant analysis (LDA), which is a supervised technique, was also performed to further analyse the scores obtained from PCA 21 . The number of the chosen PCs was the best compromise between a minimal number of PCs and classification accuracy.…”

“…HCA is an unsupervised method for obtaining information and it clusters the spectra by spectral similarity. Since HCA is a distance based method, any distance metrics such as Euclidean or Mahalanobis can be employed [21] . In the present study, the dendrogram was generated using Ward's clustering algorithm and the squared Euclidean distance.…”

“…Moreover, aqueous samples can be easily studied as water is a weak Raman scatterer and enables monitoring of biological systems in live condition. Furthermore, the level of specificity can be enhanced by combining this technique with chemometric methods 21,22 as even the most subtle changes are expected to be picked up by this technique.…”

Net Analyte Signal-based Quantitative Determination of Fusel Oil in Korean Alcoholic Beverage Using FT-NIR Spectroscopy

“…The predictor variables in spectroscopic data (particularly in the near-IR and mid-IR) are highly correlated with each other, which leads to an ill-conditioned least-squares problem (Gautam et al, 2015). Therefore, multivariate calibration methods such as PLSR and PCR must reduce the size of the predictor in latent variables, which is the most relevant information in the spectrum.…”

“…However, in PLSR, covariance between the predictor and response variables, i.e., X T Y, is subjected to singular value decomposition. This is in contrast to X T X in PCR, which obtains the latent variables used to predict the response variables (Gautam et al, 2015). In this work, we developed the PCR and PLSR models with both raw and NAS preprocessed data.…”