Siyuan Hou

Exploratory data analysis with noisy measurements

Multivariate chemical and biological data are increasingly characterized by measurement error variances that are highly heterogeneous. Such heteroscedasticity may be inherent in the measurements themselves or a consequence of data pretreatment. The presence of measurements with large error variances among more precise observations leads to problems in data analysis. For exploratory data analysis and in particular the low‐dimensional visualization of data structures, these complications can result from sources that include preprocessing, subspace estimation, and the projection of objects with erroneous measurements, as well as contamination of the projection space with unreliable samples that preclude the effective visualization of data structures that may be present. In this work, a general strategy is proposed for the exploratory data analysis of multivariate data exhibiting a high degree of non‐uniformity in measurement error variance, where estimates of the variance are available. This strategy involves three principles: (1) mitigation of the effects of noisy measurements through a preprocessing step that uses maximum likelihood principal components analysis; (2) propagation of measurement uncertainty through all steps of the procedure; and (3) incorporation of the uncertainty information into the projection of data onto the visualization subspace. To carry out this last step, a new technique, referred to as the partial transparency projection, is introduced in which the quality of measurements is interactively imbedded into the appearance of the object in the space. The advantages of this strategy are demonstrated with simulated measurements and using experimental microarray gene expression data from a yeast time course study. Copyright

Re-centered kurtosis as a projection pursuit index for multivariate data analysis

High‐dimensional data, which have become common in analytical chemistry, are often rich in information, but useful information may not be discovered without applying advanced data analysis methods. As a powerful tool for exploratory data analysis, projection pursuit (PP) is less widely used in chemistry compared with other methods such as principal component analysis (PCA), although PP often gives better results than PCA. PP does not have a uniquely defined objective function (projection index), and different statistics have been proposed as projection indices. Kurtosis has been widely employed as a projection index, and minimization of kurtosis is helpful in revealing clusters. However, this method often fails when the clusters in a data set are not balanced (i.e., present in unequal proportions). In this work, a newly defined kurtosis, referred to as “re‐centered kurtosis,” is proposed as a projection index. The theory and the optimization algorithms for the re‐centered kurtosis are developed. The utility of the PP method using the proposed re‐centered kurtosis as a projection index to reveal unbalanced clusters is demonstrated by simulated and real experimental data. Copyright

Exploration of attenuated total reflectance mid-infrared spectroscopy and multivariate calibration to measure immunoglobulin G in human sera

Immunoglobulin G (IgG) is crucial for the protection of the host from invasive pathogens. Due to its importance for human health, tools that enable the monitoring of IgG levels are highly desired. Consequently there is a need for methods to determine the IgG concentration that are simple, rapid, and inexpensive. This work explored the potential of attenuated total reflectance (ATR) infrared spectroscopy as a method to determine IgG concentrations in human serum samples. Venous blood samples were collected from adults and children, and from the umbilical cord of newborns. The serum was harvested and tested using ATR infrared spectroscopy. Partial least squares (PLS) regression provided the basis to develop the new analytical methods. Three PLS calibrations were determined: one for the combined set of the venous and umbilical cord serum samples, the second for only the umbilical cord samples, and the third for only the venous samples. The number of PLS factors was chosen by critical evaluation of Monte Carlo-based cross validation results. The predictive performance for each PLS calibration was evaluated using the Pearson correlation coefficient, scatter plot and Bland-Altman plot, and percent deviations for independent prediction sets. The repeatability was evaluated by standard deviation and relative standard deviation. The results showed that ATR infrared spectroscopy is potentially a simple, quick, and inexpensive method to measure IgG concentrations in human serum samples. The results also showed that it is possible to build a united calibration curve for the umbilical cord and the venous samples.

Measurement of serum immunoglobulin G in dairy cattle using Fourier-transform infrared spectroscopy: A reagent free approach

Simple, rapid and cost-effective methods are sought for measuring immunoglobulin G (IgG) concentrations in bovine serum, which can be applied for diagnosis of failure of transfer of passive immunity (FTPI). The aim of the present study was to investigate the potential use of Fourier-transform infrared (FTIR) spectroscopy, with partial least squares (PLS) regression, to measure IgG concentrations in bovine serum. Serum samples collected from calves and adult cows were tested in parallel by radial immunodiffusion (RID) assay and FTIR spectroscopy. The sample IgG concentrations obtained by the RID method were linked to pre-processed spectra and divided into two sets: a combined set and a test set. The combined set was used for building a calibration model, while the test set was used to assess the predictive ability of the calibration model, resulting in a root mean squared error of prediction (RMSEP) of 307.5 mg/dL. The concordance correlations between the IgG measured by RID and predicted by FTIR spectroscopy were 0.96 and 0.93 for the combined and test data sets, respectively. Analysis of the data using the Bland-Altman method did not show any evidence of systematic bias between FTIR spectroscopy and RID methods for measurement of IgG. The clinical applicability of FTIR spectroscopy for diagnosis of FTPI was evaluated using the entire data set and showed a sensitivity of 0.91 and specificity of 0.96, using RID as the reference standard. The FTIR spectroscopy method, described in the present study, demonstrates potential as a rapid and reagent-free tool for quantification of IgG in bovine serum, as an aid to diagnosis of FTPI in calves.

Use of Fourier-Transform Infrared Spectroscopy to Quantify Immunoglobulin G Concentrations in Alpaca Serum

Background Rapid, economical, and quantitative assays for measurement of camelid serum immunoglobulin G (IgG) are limited. In camelids, failure of transfer of maternal immunoglobulins has a reported prevalence of up to 20.5%. An accurate method for quantifying serum IgG concentrations is required. Objective To develop an infrared spectroscopy‐based assay for measurement of alpaca serum IgG and compare its performance to the reference standard radial immunodiffusion (RID) assay. Animals One hundred and seventy‐five privately owned, healthy alpacas. Methods Eighty‐two serum samples were collected as convenience samples during routine herd visits whereas 93 samples were recruited from a separate study. Serum IgG concentrations were determined by RID assays and midinfrared spectra were collected for each sample. Fifty samples were set aside as the test set and the remaining 125 training samples were employed to build a calibration model using partial least squares (PLS) regression with Monte Carlo cross validation to determine the optimum number of PLS factors. The predictive performance of the calibration model was evaluated by the test set. Results Correlation coefficients for the IR‐based assay were 0.93 and 0.87, respectively, for the entire data set and test set. Sensitivity in the diagnosis of failure of transfer of passive immunity (FTPI) ([IgG] <1,000 mg/dL) was 71.4% and specificity was 100% for the IR‐based method (test set) as gauged relative to the RID reference method assay. Conclusions and Clinical Importance This study indicated that infrared spectroscopy, in combination with chemometrics, is an effective method for measurement of IgG in alpaca serum.

Use of Fourier-transform infrared spectroscopy to quantify immunoglobulin G concentration and an analysis of the effect of signalment on levels in canine serum

Deficiency in immunoglobulin G (IgG) is associated with an increased susceptibility to infections in humans and animals, and changes in IgG levels occur in many disease states. In companion animals, failure of transfer of passive immunity is uncommonly diagnosed but mortality rates in puppies are high and more than 30% of these deaths are secondary to septicemia. Currently, radial immunodiffusion (RID) and enzyme-linked immunosorbent assays are the most commonly used methods for quantitative measurement of IgG in dogs. In this study, a Fourier-transform infrared spectroscopy (FTIR) assay for canine serum IgG was developed and compared to the RID assay as the reference standard. Basic signalment data and health status of the dogs were also analyzed to determine if they correlated with serum IgG concentrations based on RID results. Serum samples were collected from 207 dogs during routine hematological evaluation, and IgG concentrations determined by RID. The FTIR assay was developed using partial least squares regression analysis and its performance evaluated using RID assay as the reference test. The concordance correlation coefficient was 0.91 for the calibration model data set and 0.85 for the prediction set. A Bland-Altman plot showed a mean difference of -89 mg/dL and no systematic bias. The modified mean coefficient of variation (CV) for RID was 6.67%, and for FTIR was 18.76%. The mean serum IgG concentration using RID was 1943 ± 880 mg/dL based on the 193 dogs with complete signalment and health data. When age class, gender, breed size and disease status were analyzed by multivariable ANOVA, dogs < 2 years of age (p = 0.0004) and those classified as diseased (p = 0.03) were found to have significantly lower IgG concentrations than older and healthy dogs, respectively.

Immunoglobulin G Measurement in Blood Plasma Using Infrared Spectroscopy

A rapid, simple, and inexpensive method to measure the immunoglobulin G (IgG) concentrations in blood samples in human and veterinary medicine is highly desired. Infrared spectroscopy (coupled with chemometric manipulation of spectral data) has the advantages of simple sample preparation, rapid implementation of analysis, and low cost. Here a method that exploits infrared spectroscopy as the basis to measure IgG concentration in animal plasma samples is reported, with radial immunodiffusion (RID) used as the reference test method for partial least squares (PLS) calibration model development. Smoothed non-derivative and the second-order derivative spectra were used to develop calibration models. Various additional spectral preprocessing steps were evaluated to optimize the calibration models, and the possible benefits of using an internal standard (potassium thiocyanate [KSCN]) were investigated. Monte Carlo cross-validation was used to determine the optimal number of PLS factors, and an independent prediction set was used to test the predictive performances of provisional models. The effects of various preprocessing options (spectral smoothing, derivation, normalization, region selection, mean-centering, and standard deviation scaling) on quantification accuracy were investigated. The root mean squared error of prediction (RMSEP) for different combinations of spectra preprocessing steps was 394 ± 36 mg/dL for the non-derivative spectra and 427 ± 101 mg/dL for the second-order derivative spectra. Immunoglobulin G concentrations produced by the optimized PLS model for the non-derivative spectra (RMSEP = 352 mg/dL) were found to be stable with respect to different splits of the samples among the calibration, validation, and prediction sets. The precision of the Fourier transform infrared (FT-IR) method is found to be slightly superior to that of the RID method. The results of this work indicate that infrared spectroscopy is a promising technique for economically and rapidly determining the IgG concentrations of plasma and plasma-derived samples.

Biochemical Variation Among Normal Equine Carpal and Tarsocrural Joint Fluids are Detected by Infrared Spectral Characteristics and A Modified Approach to Linear Discriminant Analysis

Research into osteoarthritis diagnostics has evolved from traditional methods that are only useful in more advanced clinical disease, towards the discovery of biomarkers that are predictive or reflective of preclinical joint disease. The potential of Fourier transform infrared spectroscopy (FTIR) coupled with chemometrics has been demonstrated as useful for the assessment of biomolecular responses to disease. Joint fluid was collected from 105 clinically normal antebrachiocarpal (AC), midcarpal (MC) and tarsocrural (TCRL) joints. Thin films were prepared and FTIR absorbance spectra in the mid-infrared region recorded. Interferograms were signal averaged and Fourier transformed to generate spectra with a nominal resolution of 4 cm-1. Comparisons among joints were made using a novel modified method similar to linear discriminant analysis, which maximized the difference of between-group variance minus within-group variance, followed by permutation testing. Differences within animal between contralateral pairs of joints were minimal. Significant differences among AC, MC and TCRL joint fluid spectra were found. The range of biomolecular differences among these normal joints as characterized by FTIR indicates that interarticular variation within the horse needs to be considered for ongoing research, especially when utilizing within-horse joints as controls.

Detection of biochemical variations among normal equine carpal and tarsocrural joint fluids based on infrared spectral characteristics and a modified approach to partial least squares discriminant analysis

In recent years the focus of osteoarthritis (OA) research has shifted from microscopic changes in cell number and total protein parameters, to the search for biomarkers (direct or indirect molecular indicators of abnormal skeletal turnover) for joint disease in horses. Infrared (IR) spectroscopy has been established as a useful tool for the assessment of biological molecules and the biochemical response to disease. Synovial fluid was collected from 105 normal clinically normal antebrachiocarpal (AC), midcarpal (MC) and tarsocrural (TC) joints. Thin films were prepared and infrared absorbance spectra in the mi-infrared (MIR) region of 400-4000 cm recorded using a FTIR spectrometer equipped with a deuterium tryglycine sulphate detector. For each acquisition, 512 interferograms were signal averaged and Fourier transformed to generate a spectrum with nominal resolution of 4 cm-1. Spectra were preprocessed and left and right spectra compared by ANOVA. Between joint comparisons were made using a novel modified method stemming from of partial least squares discriminant analysis and projection pursuit methods that maximized the difference of between-group variance minus within-group variance. Differences within animal between left and right joints were minimal. Differences among AC, MC and TC synovial fluid spectra were significant. The finding of a broad range biomolecular differences among these joints supports indicates that interarticular variation within the horse needs to be considered. Keywords-infrared spectroscopy, osteoarthritis, synovial fluid, discriminant analysis