Multivariate pattern analysis: a method and software to reveal, quantify, and visualize predictive association patterns in multicollinear data.

BACKGROUND: Strongly multicollinear covariates, such as those typically represented in metabolomics applications, represent a challenge for multivariate regression analysis. These challenges are commonly circumvented by reducing the number of covariates to a subset of linearly independent variables, but this strategy may lead to loss of resolution and thus produce models with poorer interpretative potential. The aim of this work was to implement and illustrate a method, multivariate pattern analysis (MVPA), which can handle multivariate covariates without compromising resolution or model quality. RESULTS: MVPA has been implemented in an open-source R package of the same name, mvpa. To facilitate the usage and interpretation of complex association patterns, mvpa has also been integrated into an R shiny app, mvpaShiny, which can be accessed on www.mvpashiny.org . MVPA utilizes a general projection algorithm that embraces a diversity of possible models. The method handles multicollinear and even linear dependent covariates. MVPA separates the variance in the data into orthogonal parts within the frame of a single joint model: one part describing the relations between covariates, outcome, and explanatory variables and another part describing the "net" predictive association pattern between outcome and explanatory variables. These patterns are visualized and interpreted in variance plots and plots for pattern analysis and ranking according to variable importance. Adjustment for a linear dependent covariate is performed in three steps. First, partial least squares regression with repeated Monte Carlo resampling is used to determine the number of predictive PLS components for a model relating the covariate to the outcome. Second, postprocessing of this PLS model by target projection provided a single component expressing the predictive association pattern between the outcome and the covariate. Third, the outcome and the explanatory variables were adjusted for the covariate by using the target score in the projection algorithm to obtain "net" data. We illustrate the main features of MVPA by investigating the partial mediation of a linearly dependent metabolomics descriptor on the association pattern between a measure of insulin resistance and lifestyle-related factors. CONCLUSIONS: Our method and implementation in R extend the range of possible analyses and visualizations that can be performed for complex multivariate data structures. The R packages are available on github.com/liningtonlab/mvpa and github.com/liningtonlab/mvpaShiny.

An approach to assess and adjust for the influence of multicollinear covariates on metabolomics association patterns-applied to a study of the associations between a comprehensive lipoprotein profile and the homeostatic model assessment of insulin resistance.

INTRODUCTION: Comprehensive lipoprotein profiling using proton nuclear magnetic resonance (NMR) spectroscopy of serum represents an alternative to the homeostatic model assessment of insulin resistance (HOMA-IR). Both adiposity and physical (in)activity associate to insulin resistance, but quantification of the influence of these two lifestyle related factors on the association pattern of HOMA-IR to lipoproteins suffers from lack of appropriate methods to handle multicollinear covariates. OBJECTIVES: We aimed at (i) developing an approach for assessment and adjustment of the influence of multicollinear and even linear dependent covariates on regression models, and (ii) to use this approach to examine the influence of adiposity and physical activity on the association pattern between HOMA-IR and the lipoprotein profile. METHODS: For 841 children, lipoprotein profiles were obtained from serum proton NMR and physical activity (PA) intensity profiles from accelerometry. Adiposity was measured as body mass index, the ratio of waist circumference to height, and skinfold thickness. Target projections were used to assess and isolate the influence of adiposity and PA on the association pattern of HOMA-IR to the lipoproteins. RESULTS: Adiposity explained just over 50% of the association pattern of HOMA-IR to the lipoproteins with strongest influence on high-density lipoprotein features. The influence of PA was mainly attributed to a strong inverse association between adiposity and moderate and high-intensity physical activity. CONCLUSION: The presented covariate projection approach to obtain net association patterns, made it possible to quantify and interpret the influence of adiposity and physical (in)activity on the association pattern of HOMA-IR to the lipoprotein features.

Physical activity intensity profiles associated with cardiometabolic risk in middle-aged to older men and women.

Accelerometers provide detailed data about physical activity (PA) across the full intensity spectrum. However, when examining associations with health, results are often aggregated to only a few summary measures [e.g. time spent "sedentary" or "moderate-to-vigorous" intensity PA]. Using multivariate pattern analysis, which can handle collinear exposure variables, we examined associations between the full PA intensity spectrum and cardiometabolic risk (CMR) in a population-based sample of middle-aged to older adults. Participants (n = 3660; mean ± SD age = 69 ± 8y and BMI = 26.7 ± 4.2 kg/m2; 55% female) from the EPIC-Norfolk study (UK) with valid accelerometry (ActiGraph-GT1M) data were included. We used multivariate pattern analysis with partial least squares regression to examine cross-sectional multivariate associations (r) across the full PA intensity spectrum [minutes/day at 0-5000 counts-per-minute (cpm); 5 s epoch] with a continuous CMR score (reflecting waist, blood pressure, lipid, and glucose metabolism). Models were sex-stratified and adjusted for potential confounders. There was a positive (detrimental) association between PA and CMR at 0-12 cpm (maximally-adjusted r = 0.08 (95%CI 0.06-0.10). PA was negatively (favourably) associated with CMR at all intensities above 13 cpm ranging between r = -0.09 (0.07-0.12) at 800-999 cpm and r = -0.14 (0.11-0.16) at 75-99 and 4000-4999 cpm. The strongest favourable associations were from 50 to 800 cpm (r = 0.10-0.12) in men, but from ≥2500 cpm (r = 0.18-0.20) in women; with higher proportions of model explained variance for women (R2 = 7.4% vs. 2.3%). Most of the PA intensity spectrum was beneficially associated with CMR in middle-aged to older adults, even at intensities lower than what has traditionally been considered "sedentary" or "light-intensity" activity. This supports encouragement of PA at almost any intensity in this age-group.

Opportunities and Limitations for Untargeted Mass Spectrometry Metabolomics to Identify Biologically Active Constituents in Complex Natural Product Mixtures.

Compounds derived from natural sources represent the majority of small-molecule drugs utilized today. Plants, owing to their complex biosynthetic pathways, are poised to synthesize diverse secondary metabolites that selectively target biological macromolecules. Despite the vast chemical landscape of botanicals, drug discovery programs from these sources have diminished due to the costly and time-consuming nature of standard practices and high rates of compound rediscovery. Untargeted metabolomics approaches that integrate biological and chemical data sets potentially enable the prediction of active constituents early in the fractionation process. However, data acquisition and data processing parameters may have major impacts on the success of models produced. Using an inactive botanical mixture spiked with known antimicrobial compounds, untargeted mass spectrometry-based metabolomics data were combined with bioactivity data to produce selectivity ratio models subjected to a variety of data acquisition and data processing parameters. Selectivity ratio models were used to identify active constituents that were intentionally added to the mixture, along with an additional antimicrobial compound, randainal (5), which was masked by the presence of antagonists in the mixture. These studies found that data-processing approaches, particularly data transformation and model simplification tools using a variance cutoff, had significant impacts on the models produced, either masking or enhancing the ability to detect active constituents in samples. The current study highlights the importance of the data processing step for obtaining reliable information from metabolomics models and demonstrates the strengths and limitations of selectivity ratio analysis to comprehensively assess complex botanical mixtures.

Biochemometrics to Identify Synergists and Additives from Botanical Medicines: A Case Study with Hydrastis canadensis (Goldenseal).

A critical challenge in the study of botanical natural products is the difficulty of identifying multiple compounds that may contribute additively, synergistically, or antagonistically to biological activity. Herein, it is demonstrated how combining untargeted metabolomics with synergy-directed fractionation can be effective toward accomplishing this goal. To demonstrate this approach, an extract of the botanical goldenseal ( Hydrastis canadensis) was fractionated and tested for its ability to enhance the antimicrobial activity of the alkaloid berberine (4) against the pathogenic bacterium Staphylococcus aureus. Bioassay data were combined with untargeted mass spectrometry-based metabolomics data sets (biochemometrics) to produce selectivity ratio (SR) plots, which visually show which extract components are most strongly associated with the biological effect. Using this approach, the new flavonoid 3,3'-dihydroxy-5,7,4'-trimethoxy-6,8- C-dimethylflavone (29) was identified, as were several flavonoids known to be active. When tested in combination with 4, 29 lowered the IC50 of 4 from 132.2 ± 1.1 µM to 91.5 ± 1.1 µM. In isolation, 29 did not demonstrate antimicrobial activity. The current study highlights the importance of fractionation when utilizing metabolomics for identifying bioactive components from botanical extracts and demonstrates the power of SR plots to help merge and interpret complex biological and chemical data sets.

Integration of Biochemometrics and Molecular Networking to Identify Antimicrobials in Angelica keiskei.

Botanical medicines have been utilized for centuries, but it remains challenging to identify bioactive constituents from complex botanical extracts. Bioassay-guided fractionation is often biased toward abundant or easily isolatable compounds. To comprehensively evaluate active botanical mixtures, methods that allow for the prioritization of active compounds are needed. To this end, a method integrating bioassay-guided fractionation, biochemometric selectivity ratio analysis, and molecular networking was devised and applied to Angelica keiskei to comprehensively evaluate its antimicrobial activity against Staphylococcus aureus. This approach enabled the identification of putative active constituents early in the fractionation process and provided structural information for these compounds. A subset of chalcone analogs were prioritized for isolation, yielding 4-hydroxyderricin (1, minimal inhibitory concentration [MIC] ≤ 4.6 µM, IC50 = 2.0 µM), xanthoangelol (2, MIC ≤ 4.0 µM, IC50 = 2.3) and xanthoangelol K (4, IC50 = 168 µM). This approach allowed for the identification of a low-abundance compound (xanthoangelol K) that has not been previously reported to possess antimicrobial activity and facilitated a more comprehensive understanding of the compounds responsible for A. keiskei's antimicrobial activity.

Comparison of Metabolomics Approaches for Evaluating the Variability of Complex Botanical Preparations: Green Tea (Camellia sinensis) as a Case Study.

A challenge that must be addressed when conducting studies with complex natural products is how to evaluate their complexity and variability. Traditional methods of quantifying a single or a small range of metabolites may not capture the full chemical complexity of multiple samples. Different metabolomics approaches were evaluated to discern how they facilitated comparison of the chemical composition of commercial green tea [Camellia sinensis (L.) Kuntze] products, with the goal of capturing the variability of commercially used products and selecting representative products for in vitro or clinical evaluation. Three metabolomic-related methods-untargeted ultraperformance liquid chromatography-mass spectrometry (UPLC-MS), targeted UPLC-MS, and untargeted, quantitative 1HNMR-were employed to characterize 34 commercially available green tea samples. Of these methods, untargeted UPLC-MS was most effective at discriminating between green tea, green tea supplement, and non-green-tea products. A method using reproduced correlation coefficients calculated from principal component analysis models was developed to quantitatively compare differences among samples. The obtained results demonstrated the utility of metabolomics employing UPLC-MS data for evaluating similarities and differences between complex botanical products.

Effects of physical activity on schoolchildren's academic performance: The Active Smarter Kids (ASK) cluster-randomized controlled trial.

OBJECTIVE: To investigate the effect of a seven-month, school-based cluster-randomized controlled trial on academic performance in 10-year-old children. METHODS: In total, 1129 fifth-grade children from 57 elementary schools in Sogn og Fjordane County, Norway, were cluster-randomized by school either to the intervention group or to the control group. The children in the 28 intervention schools participated in a physical activity intervention between November 2014 and June 2015 consisting of three components: 1) 90min/week of physically active educational lessons mainly carried out in the school playground; 2) 5min/day of physical activity breaks during classroom lessons; 3) 10min/day physical activity homework. Academic performance in numeracy, reading and English was measured using standardized Norwegian national tests. Physical activity was measured objectively by accelerometry. RESULTS: We found no effect of the intervention on academic performance in primary analyses (standardized difference 0.01-0.06, p>0.358). Subgroup analyses, however, revealed a favorable intervention effect for those who performed the poorest at baseline (lowest tertile) for numeracy (p=0.005 for the subgroup∗group interaction), compared to controls (standardized difference 0.62, 95% CI 0.19-1.07). CONCLUSIONS: This large, rigorously conducted cluster RCT in 10-year-old children supports the notion that there is still inadequate evidence to conclude that increased physical activity in school enhances academic achievement in all children. Still, combining physical activity and learning seems a viable model to stimulate learning in those academically weakest schoolchildren.

Biochemometrics for Natural Products Research: Comparison of Data Analysis Approaches and Application to Identification of Bioactive Compounds.

A central challenge of natural products research is assigning bioactive compounds from complex mixtures. The gold standard approach to address this challenge, bioassay-guided fractionation, is often biased toward abundant, rather than bioactive, mixture components. This study evaluated the combination of bioassay-guided fractionation with untargeted metabolite profiling to improve active component identification early in the fractionation process. Key to this methodology was statistical modeling of the integrated biological and chemical data sets (biochemometric analysis). Three data analysis approaches for biochemometric analysis were compared, namely, partial least-squares loading vectors, S-plots, and the selectivity ratio. Extracts from the endophytic fungi Alternaria sp. and Pyrenochaeta sp. with antimicrobial activity against Staphylococcus aureus served as test cases. Biochemometric analysis incorporating the selectivity ratio performed best in identifying bioactive ions from these extracts early in the fractionation process, yielding altersetin (3, MIC 0.23 µg/mL) and macrosphelide A (4, MIC 75 µg/mL) as antibacterial constituents from Alternaria sp. and Pyrenochaeta sp., respectively. This study demonstrates the potential of biochemometrics coupled with bioassay-guided fractionation to identify bioactive mixture components. A benefit of this approach is the ability to integrate multiple stages of fractionation and bioassay data into a single analysis.

Active Smarter Kids (ASK): Rationale and design of a cluster-randomized controlled trial investigating the effects of daily physical activity on children's academic performance and risk factors for non-communicable diseases.

BACKGROUND: Evidence is emerging from school-based studies that physical activity might favorably affect children's academic performance. However, there is a need for high-quality studies to support this. Therefore, the main objective of the Active Smarter Kids (ASK) study is to investigate the effect of daily physical activity on children's academic performance. Because of the complexity of the relation between physical activity and academic performance it is important to identify mediating and moderating variables such as cognitive function, fitness, adiposity, motor skills and quality of life (QoL). Further, there are global concerns regarding the high prevalence of lifestyle-related non-communicable diseases (NCDs). The best means to address this challenge could be through primary prevention. Physical activity is known to play a key role in preventing a host of NCDs. Therefore, we investigated as a secondary objective the effect of the intervention on risk factors related to NCDs. The purpose of this paper is to describe the design of the ASK study, the ASK intervention as well as the scope and details of the methods we adopted to evaluate the effect of the ASK intervention on 5 (th) grade children. METHODS & DESIGN: The ASK study is a cluster randomized controlled trial that includes 1145 fifth graders (aged 10 years) from 57 schools (28 intervention schools; 29 control schools) in Sogn and Fjordane County, Norway. This represents 95.3 % of total possible recruitment. Children in all 57 participating schools took part in a curriculum-prescribed physical activity intervention (90 min/week of physical education (PE) and 45 min/week physical activity, in total; 135 min/week). In addition, children from intervention schools also participated in the ASK intervention model (165 min/week), i.e. a total of 300 min/week of physical activity/PE. The ASK study was implemented over 7 months, from November 2014 to June 2015. We assessed academic performance in reading, numeracy and English using Norwegian National tests delivered by The Norwegian Directorate for Education and Training. We assessed physical activity objectively at baseline, midpoint and at the end of the intervention. All other variables were measured at baseline and post-intervention. In addition, we used qualitative methodologies to obtain an in-depth understanding of children's embodied experiences and pedagogical processes taking place during the intervention. DISCUSSION: If successful, ASK could provide strong evidence of a relation between physical activity and academic performance that could potentially inform the process of learning in elementary schools. Schools might also be identified as effective settings for large scale public health initiatives for the prevention of NCDs. TRIAL REGISTRATION: Clinicaltrials.gov ID nr: NCT02132494 . Date of registration, 6(th) of May, 2014.

Variation in Phenolic Chemistry in Zostera marina Seagrass Along Environmental Gradients.

Chemical ecology has been suggested as a less time-consuming and more cost-efficient monitoring tool of seagrass ecosystems than traditional methods. Phenolic chemistry in Zostera marina samples was analyzed against latitude, sea depth, sample position within a seagrass meadow (periphery or center) and wave exposure. Multivariate data analysis showed that rosmarinic acid correlated moderately positively with depth, while the flavonoids had an overall strong negative correlation with increasing depth-possibly reflecting lack of stress-induced conditions with increasing depth, rather than a different response to light conditions. At a molecular level, the flavonoids were separated into two groups; one group is well described by the variables of depth and wave exposure, and the other group that was not well described by these variables-the latter may reflect biosynthetic dependencies or other unrevealed factors. A higher flavonoid/rosmarinic acid ratio was seen in the periphery of a seagrass meadow, while the contrary ratio was seen in the center. This may reflect higher plant stress in the periphery of a meadow, and the flavonoid/rosmarinic acid ratio may provide a possible molecular index of seagrass ecosystem health. Further studies are needed before the full potential of using variation in phenolic chemistry as a seagrass ecosystem monitoring tool is established.

A multivariate approach to reveal biomarker signatures for disease classification: application to mass spectral profiles of cerebrospinal fluid from patients with multiple sclerosis.

Mass spectral profiles from cerebrospinal fluid (CSF) are used as input to a novel multivariate approach to select features responsible for the separation of patients with multiple sclerosis (MS) from control groups. Our targeted statistical approach makes it possible to systematically remove features in the spectral fingerprints masking the components expressing the disease pattern. The low molecular weight CSF proteome from 54 patients with MS and a range of other neurological diseases (OND), as well as neurological healthy controls (NHC), is analyzed in replicates using mass spectral profiling. Statistically validated partial least-squares discriminant analysis (PLS-DA) models are created as a first step to separate the groups. Using the group membership as a target, the most discriminatory projection in the multivariate space spanned by the spectral profiles is revealed. From the resulting target-projected component, the spectral regions most significantly contributing to group separation are identified using the nonparametric discriminating variable (DIVA) test together with the so-called selectivity ratio (SR) plot. Our approach is general and can be applied for other diseases and instrumental techniques as well.

Composite score analysis for unsupervised comparison and network visualization of metabolomics data.

Metabolomics-based approaches are becoming increasingly popular to interrogate the chemical basis for phenotypic differences in biological systems. Successful metabolomics studies employ multivariate data analysis to compare large and highly complex datasets. A primary tool for unsupervised statistical analyses, principal component analysis (PCA), relies on the selection of a subsection of a maximum of three components from a larger model to visually represent similarity. The use of only three principal components limits the comprehensiveness of the model and can mask discrimination between samples. We have developed a new statistical metric, the composite score (CS), as a univariate statistic that incorporates multiple principal components to calculate a correlation matrix that enables quantitative comparisons of sample similarity between samples within one dataset based upon measured metabolome profiles. Composite score values were tabulated using profiles of complex extracts of dietary supplements from the plant Hydrastis canadensis (goldenseal) as a case study. Several outliers were unambiguously identified, and a PCA composite score network was developed to provide a graphical representation of the composite score matrix. Comparison with visualization using PCA score plots or dendrograms from hierarchical clustering analysis (HCA) demonstrates the utility of the composite score to as a tool for metabolomics studies that seek to quantify similarity among samples. An R-script for the calculation of composite score has been made available.

Discriminating variable test and selectivity ratio plot: quantitative tools for interpretation and variable (biomarker) selection in complex spectral or chromatographic profiles.

The discriminating variable (DIVA) test and the selectivity ratio (SR) plot are developed as quantitative tools for revealing the variables in spectral or chromatographic profiles discriminating best between two groups of samples. The SR plot is visually similar to a spectrum or a chromatogram, but with the most intense regions corresponding to the most discriminating variables. Thus, the variables with highest SR represent the variables most important for interpretation of differences between groups. Regions with variables that are positively or negatively correlated to each other are displayed as corresponding negative and positive regions in the SR plot. The nonparametric DIVA test is designed for connecting SR to discriminatory ability of a variable quantified as probability for correct classification. A mean probability for a certain SR range is calculated as the mean correct classification rate (MCCR) for all variables in the same SR interval. The MCCR is thus similar to a mean sensitivity in each SR interval. In addition to the ranking of all variables according to their discriminatory ability provided by the SR plot, the DIVA test connects a probability measure to each SR interval. Thus, the DIVA test makes it possible to objectively define thresholds corresponding to mean probability levels in the SR plot and provides a quantitative means to select discriminating variables. In order to validate the approach, samples of untreated cerebrospinal fluid (CSF) and samples spiked with a multicomponent peptide standard were analyzed by matrix-assisted laser desorption ionization (MALDI) mass spectrometry. The differences in the multivariate spectral profiles of the two groups were revealed using partial least-squares discriminant analysis (PLS-DA) followed by target projection (TP). The most discriminating mass-to-charge (m/z) regions were revealed by calculating the ratio of explained to unexplained variance for each m/z number on the target-projected component and displaying this measure in SR plots with quantitative boundaries determined from the DIVA test. The results are compared to some established methods for variable selection.

Recipe for uncovering the bioactive components in herbal medicine.

Using whole chromatographic profiles and measurements of total bioactivity as input, a quantitative pattern-activity relationship (QPAR) approach is proposed as a general method for providing two pieces of crucial information about complex bioactive mixtures available: (i) a model for predicting total bioactivity from the chromatographic fingerprint and (ii) the features in the chromatographic profile responsible for the bioactivity. While the first piece of information is already available through existing approaches, the second one results from our ability to remove dominant features in the chromatographic fingerprints which mask the components specifically related to pharmacological activity. Our targeted approach makes information about bioactivity available at the molecular level and provides possibilities for assessment of herbal medicine (HM) possible beyond just authentication and total bioactivity. As an example, the antioxidant property of the HM Radix Puerariae lobatae is measured through its reducing power toward a ferric ion complex. A partial least-squares (PLS) model is created to predict the antioxidant activity from the chromatographic fingerprint. Using the antioxidant activity as a target, the most discriminatory projection in the multivariate space spanned by the chromatographic profiles is revealed. From this target-projected component, the chromatographic regions most strongly connected to antioxidant activity are identified using the so-called selectivity ratio (SR) plot. The results are validated by prediction of samples not included in the modeling step.

Hierarchical cluster analysis of technical replicates to identify interferents in untargeted mass spectrometry metabolomics.

Mass spectral data sets often contain experimental artefacts, and data filtering prior to statistical analysis is crucial to extract reliable information. This is particularly true in untargeted metabolomics analyses, where the analyte(s) of interest are not known a priori. It is often assumed that chemical interferents (i.e. solvent contaminants such as plasticizers) are consistent across samples, and can be removed by background subtraction from blank injections. On the contrary, it is shown here that chemical contaminants may vary in abundance across each injection, potentially leading to their misidentification as relevant sample components. With this metabolomics study, we demonstrate the effectiveness of hierarchical cluster analysis (HCA) of replicate injections (technical replicates) as a methodology to identify chemical interferents and reduce their contaminating contribution to metabolomics models. Pools of metabolites with varying complexity were prepared from the botanical Angelica keiskei Koidzumi and spiked with known metabolites. Each set of pools was analyzed in triplicate and at multiple concentrations using ultraperformance liquid chromatography coupled to mass spectrometry (UPLC-MS). Before filtering, HCA failed to cluster replicates in the data sets. To identify contaminant peaks, we developed a filtering process that evaluated the relative peak area variance of each variable within triplicate injections. These interferent peaks were found across all samples, but did not show consistent peak area from injection to injection, even when evaluating the same chemical sample. This filtering process identified 128 ions that appear to originate from the UPLC-MS system. Data sets collected for a high number of pools with comparatively simple chemical composition were highly influenced by these chemical interferents, as were samples that were analyzed at a low concentration. When chemical interferent masses were removed, technical replicates clustered in all data sets. This work highlights the importance of technical replication in mass spectrometry-based studies, and presents a new application of HCA as a tool for evaluating the effectiveness of data filtering prior to statistical analysis.

Toxicological evaluation of complex mixtures by pattern recognition: correlating chemical fingerprints to mutagenicity.

We describe the use of pattern recognition and multivariate regression in the assessment of complex mixtures by correlating chemical fingerprints to the mutagenicity of the mixtures. Mixtures were 20 organic extracts of exhaust particles, each containing 102-170 individual compounds such as polycyclic aromatic hydrocarbons (PAHs), nitro-PAHs, oxy-PAHs, and saturated hydrocarbons. Mixtures were characterized by full-scan GC-MS (gas chromatography-mass spectrometry). Data were resolved into peaks and spectra for individual compounds by an automated curve resolution procedure. Resolved chromatograms were integrated, resulting in a predictor matrix that was used as input to a principal component analysis to evaluate similarities between mixtures (i.e., classification). Furthermore, partial least-squares projections to latent structures were used to correlate the GC-MS data to mutagenicity, as measured in the Ames Salmonella assay (i.e., calibration). The best model (high r2 and Q2) identifies the variables that co-vary with the observed mutagenicity. These variables may subsequently be identified in more detail. Furthermore, the regression model can be used to predict mutagenicity from GC-MS chromatograms of other organic extracts. We emphasize that both chemical fingerprints as well as detailed data on composition can be used in pattern recognition.

Toxicological evaluation of complex mixtures: fingerprinting and multivariate analysis.

The present paper describes a strategy for toxicological evaluation of complex mixtures based on chemical "fingerprinting" followed by pattern recognition (multivariate data analysis). The purpose is to correlate chemical fingerprints to measured toxicological endpoints, identify all major contributors to toxicity, and predict toxicity of additional mixtures. The strategy is illustrated with organic extracts of exhaust particles which are characterized by full scan gas chromatography-mass spectrometry (GC-MS). The complex GC-MS data are resolved into peaks and spectra for individual compounds using an automated curve resolution procedure. Projections to latent structures (PLS) is used for the regression modeling to correlate the GC-MS data to the measured responses; mutagenicity in the Ames Salmonella assay. The regression model identifies those peaks that co-vary with the observed mutagenicity. These peaks may be identified chemically from their spectra. Furthermore, the regression model can be used to predict mutagenicity from GC-MS chromatograms of additional samples.

Investigation of serum protein profiles in scrapie infected sheep by means of SELDI-TOF-MS and multivariate data analysis.

BACKGROUND: Classical scrapie in sheep is a fatal neurodegenerative disease associated with the conversion PrPC to PrPSc. Much is known about genetic susceptibility, uptake and dissemination of PrPSc in the body, but many aspects of prion diseases are still unknown. Different proteomic techniques have been used during the last decade to investigate differences in protein profiles between affected animals and healthy controls. We have investigated the protein profiles in serum of sheep with scrapie and healthy controls by SELDI-TOF-MS and LC-MS/MS. Latent Variable methods such as Principal Component Analysis, Partial Least Squares-Discriminant Analysis and Target Projection methods were used to describe the MS data. RESULTS: The serum proteomic profiles showed variable differences between the groups both throughout the incubation period and at the clinical end stage of scrapie. At the end stage, the target projection model separated the two groups with a sensitivity of 97.8%, and serum amyloid A was identified as one of the protein peaks that differed significantly between the groups. CONCLUSIONS: At the clinical end stage of classical scrapie, ten SELDI peaks significantly discriminated the scrapie group from the healthy controls. During the non-clinical incubation period, individual SELDI peaks were differently expressed between the groups at different time points. Investigations of differences in -omic profiles can contribute to new insights into the underlying disease processes and pathways, and advance our understanding of prion diseases, but comparison and validation across laboratories is difficult and challenging.