Principal Component Analysis with Linear and Quadratic Discriminant Analysis for Identification of Cancer Samples Based on Mass Spectrometry

Equine Veterinary Journal

et al. 2019

Summary Background Reliable and validated biomarkers for osteoarthritis (OA) are currently lacking. Objectives To develop an accurate and minimally invasive method to assess OA‐affected horses and provide potential spectral markers indicative of disease. Study design Observational, cross‐sectional study. Methods Our cohort consisted of 15 horses with OA and 48 without clinical signs of the disease, which were used as controls. Attenuated total reflection Fourier‐transform infrared (ATR‐FTIR) spectroscopy was used to investigate serum samples (50 μL) collected from these horses. Spectral processing and multivariate analysis revealed differences and similarities, allowing for detection of spectral biomarkers that discriminated between the two cohorts. A supervised classification algorithm, namely principal component analysis coupled with quadratic discriminant analysis (PCA‐QDA), was applied to evaluate the diagnostic accuracy. Results Segregation between the two different cohorts, OA‐affected and controls, was achieved with 100% sensitivity and specificity. The six most discriminatory peaks were attributed to proteins and lipids. Four of the spectral peaks were elevated in OA horses, which could be potentially due to an increase in lipids, protein expression levels and collagen, all of which have been previously reported in OA. Two peaks were found decreased and were tentatively assigned to the reduction of proteoglycan content that is observed during OA. Main limitations The control group had a wide range of ages and breeds. Presymptomatic OA cases were not included. Therefore, it remains unknown whether this test could also be used as an early diagnostic tool. Conclusions This spectrochemical approach could provide an accurate and cost‐effective blood test, facilitating point‐of‐care diagnosis of equine OA.

L_{ik} = {()(, t_{i} - {\bar{t}}_{k})}^{normalT} {boldC}_{pooled}^{- 1} ({boldt}_{i} - {\overset{boldt}{true}}_{k})

Q_{ik} = {()(, t_{i} - {\bar{t}}_{k})}^{normalT} {boldC}_{k}^{- 1} ({boldt}_{i} - {\overset{boldt}{true}}_{k})

where t i represents the PCA scores for sample i ;

{\bar{t}}_{k}

the mean PCA scores array for class k ; C pooled the pooled covariance matrix between the classes; and C k the variance‐covariance matrix for class k .…”

Section: Methodsmentioning

confidence: 99%

“…In PCA-LDA or PCA-QDA, the PCA scores T are used as input variables to a linear or quadratic discriminant classifier. PCA-LDA (L ik ) and PCA-QDA (Q ik ) classification scores are calculated in a non-Bayesian form as follows [7,8]:…”

Section: Methodsmentioning

confidence: 99%

Attenuated total reflection Fourier‐transform infrared (ATR‐FTIR) spectroscopy to diagnose osteoarthritis in equine serum

Paraskevaidi

Hook

Equine Veterinary Journal

et al. 2019

“…After these decompositions, the scores matrix from PARAFAC and Tucker3 are used as input variables for LDA and QDA algorithms. LDA and QDA classification scores can be calculated in a non-Bayesian form using the Mahalanobis distance as follows [10,11]:…”

Section: Theorymentioning

confidence: 99%

“…Discriminant analysis (DA) is a supervised classification technique employed for differentiating classes based on a Mahalanobis distance calculation [10,11]. DA can be divided into linear discriminant analysis (LDA) or quadratic discriminant analysis (QDA).…”

Section: Introductionmentioning

confidence: 99%

“…In LDA, the variance structures of the classes being analysed are considered similar, therefore the discriminant function is calculated using a pooled variance-covariance matrix among the classes. However, in QDA, each class is considered to have a different variance structure; therefore, the discriminant function is calculated using the variance-covariance matrix for each class individually [11]. This property increases the classification performance of QDA over LDA when classes exhibiting large within-category variances are being analysed.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

TTWD-DA: A MATLAB toolbox for discriminant analysis based on trilinear three-way data

Chemometrics and Intelligent Laboratory Systems

Lima

Martin

2019

Three-way trilinear data is increasingly used in chemical and biochemical applications. This type of data is composed of three-way structures representing two different signal responses and one sample dimension distributed among a 3D structure, such as the data represented by fluorescence excitation emission matrices (EMMs), spectral-pH responses, spectral-kinetic responses, spectral-electric potential responses, among others. Herein, we describe a new MATLAB toolbox for classification of trilinear three-way data using discriminant analysis techniques (linear discriminant analysis [LDA], quadratic discriminant analysis [QDA], and partial least squares discriminant analysis [PLS-DA]), termed "TTWD-DA". These discrimination techniques were coupled to multivariate deconvolution techniques by means of parallel factor analysis (PARAFAC) and Tucker3 algorithm. The toolbox is based on a user-friendly graphical interface, where these algorithms can be easily applied. Also, as output, multiple figures of merit are automatically calculated, such as accuracy, sensitivity and specificity. This software is free available online.

Spectrochemical determination of effects on rat liver of binary exposure to benzo[a]pyrene and 2,2′,4,4′‐tetrabromodiphenyl ether

Liu

Luo

J of Applied Toxicology

et al. 2021

Benzo[a]pyrene (B[a]P) and polybrominated diphenyl ethers (PBDEs) are persistent environmental contaminants. The effects in organisms of exposures to binary mixtures of such contaminants remain obscure. Attenuated total reflection Fourier‐transform infrared (ATR‐FTIR) spectroscopy is a label‐free, non‐destructive analytical technique allowing spectrochemical analysis of macromolecular components, and alterations thereof, within tissue samples. Herein, we employed ATR‐FTIR spectroscopy to identify biomolecular changes in rat liver post‐exposure to B[a]P and BDE‐47 (2,2′,4,4′‐tetrabromodiphenyl ether) congener mixtures. Our results demonstrate that significant separation occurs between spectra of tissue samples derived from control versus exposure categories (accuracy = 87%; sensitivity = 95%; specificity = 79%). Additionally, there is significant spectral separation between exposed categories (accuracy = 91%; sensitivity = 98%; specificity = 90%). Segregation between control and all exposure categories were primarily associated with wavenumbers ranging from 1600 to 1700 cm−1. B[a]P and BDE‐47 alone, or in combination, induces liver damage in female rats. However, it is suggested that binary exposure apparently attenuates the toxic effects in rat liver of the individual contaminants. This is supported by morphological observations of liver tissue architecture on hematoxylin and eosin (H&E)‐stained liver sections. Such observations highlight the difficulties in predicting the endpoint effects in target tissues of exposures to mixtures of environmental contaminants.