Wastewater samples from a Swedish chemi-thermo-mechanical pulp (CTMP) mill collected at different... more Wastewater samples from a Swedish chemi-thermo-mechanical pulp (CTMP) mill collected at different purification stages in a wastewater treatment plant (WWTP) were analyzed with an amperometric enzyme-based biosensor array in a flow-injection system. In order to resolve the complex composition of the wastewater, the array consists of several sensing elements which yield a multidimensional response. We used principal component analysis (PCA) to decompose the array's responses, and found that wastewater with different degrees of pollution can be differentiated. With the help of partial least squares regression (PLS-R), we could link the sensor responses to the Microtox® toxicity parameter, as well as to global organic pollution parameters (COD, BOD, and TOC). From investigating the influences of individual sensors in the array, it was found that the best models were in most cases obtained when all sensors in the array were included in the PLS-R model. We find that fast simultaneous determination of several global environmental parameters characterizing wastewaters is possible with this kind of biosensor array, in particular because of the link between the sensor responses and the biological effect onto the ecosystem into which the wastewater would be released. In conjunction with multivariate data analysis tools, there is strong potential to reduce the total time until a result is yielded from days to a few minutes.
The extremely high correlation between variables that is an inherent feature of near infrared (NI... more The extremely high correlation between variables that is an inherent feature of near infrared (NIR) spectroscopic data makes multivariate projection methods suitable as data analytical tools. Projection-based methods such as principal component analysis (PCA)1, partial least squares (PLS)2 regression and PLSdiscriminant analysis (PLS-DA)3 or soft independent modelling of class analogy (SIMCA)4 are therefore often used for data exploration, regression and classification respectively. In classical statistics, the canonical variates analysis (CVA) method was introduced in the 1930s by Fisher in order to optimise the group discriminative direction in multivariate space. CVA is an excellent method for estimating discriminative directions, but in contrast to e.g. PCA, the CVA method breaks down if the number of variables is larger than the number of samples and/or if the variables are highly correlated. In this respect, CVA is comparable to multiple linear regression (MLR), which suffers ...
Background/aims: Spectroscopy on human skin is a field that is being adopted increasingly because... more Background/aims: Spectroscopy on human skin is a field that is being adopted increasingly because of its rapidity and high reproducibility. Infrared reflectance (IR), near-infrared reflectance (NIR), and fluorescence spectroscopy have previously been applied to human skin in vivo to compare healthy and sick skin, including skin cancer, atopy, and leprosy. Exploratory data analysis/chemometrics is a tool for evaluating multivariate data such as spectroscopic measurements. The objective of this study was to explore the spectral variance spanned by people with normal integument, and to demonstrate the advantages of multivariate analysis to skin research. Methods: IR, NIR and fluorescence spectroscopy have been carried out in vivo on 216 volunteers' forearms before and after four tape strippings. The subjects were asked to fill in a questionnaire regarding factors suspected to influence the measurement results. Principal Component Analysis (PCA) was used to investigate whether the population can be divided into groups on the basis of their skin chemistry. Unless otherwise stated, the results are from the measurements prior to stripping. Results: In contrast to IR and fluorescence spectra, NIR spectra proved able to detect gender differences. By use of PCA, classifications on male and female subjects were observed from the IR and NIR measurements, and as an indication from the fluorescence measurements. The NIR and fluorescence measurements varied between elderly and young subjects. The largest variance in the fluorescence landscapes was seen between pigmented and non-pigmented skin. No connection was found between the spectroscopic measurements and smoking or drinking habits. Conclusions: Future spectroscopic skin investigations should be balanced as regards to gender and age, as these can possibly affect the measurement results. Chemometrics proved to be superior to traditional attempts of interpreting the spectra.
Dairy process monitoring by application of multivariate curve resolution using alternating least ... more Dairy process monitoring by application of multivariate curve resolution using alternating least squares is presented. Alternating least squares was used for resolving Fourier transform infrared spectral data from a dairy batch process in which lactose is enzymatically hydrolyzed to glucose and galactose. It was possible to extract four compounds (fat, lactose, and two other sugar components) from the spectral data obtained from nine process runs. Subsequently, the pure spectra obtained in this way were used to monitor the content of these compounds in two new process runs. In this way, alternating least squares made it possible to follow the hydrolysis process by Fourier transform infrared spectroscopy without the need for reference analyses. When the results were correlated to reference results for lactose, the accuracy was similar to that obtained when a partial least squares regression was performed on the same data; lactose correlation was 0.980 when alternating least squares was used and was 0.987 when partial least squares was used.
Wastewater samples from a Swedish chemi-thermo-mechanical pulp (CTMP) mill collected at different... more Wastewater samples from a Swedish chemi-thermo-mechanical pulp (CTMP) mill collected at different purification stages in a wastewater treatment plant (WWTP) were analyzed with an amperometric enzyme-based biosensor array in a flow-injection system. In order to resolve the complex composition of the wastewater, the array consists of several sensing elements which yield a multidimensional response. We used principal component analysis (PCA) to decompose the array's responses, and found that wastewater with different degrees of pollution can be differentiated. With the help of partial least squares regression (PLS-R), we could link the sensor responses to the Microtox® toxicity parameter, as well as to global organic pollution parameters (COD, BOD, and TOC). From investigating the influences of individual sensors in the array, it was found that the best models were in most cases obtained when all sensors in the array were included in the PLS-R model. We find that fast simultaneous determination of several global environmental parameters characterizing wastewaters is possible with this kind of biosensor array, in particular because of the link between the sensor responses and the biological effect onto the ecosystem into which the wastewater would be released. In conjunction with multivariate data analysis tools, there is strong potential to reduce the total time until a result is yielded from days to a few minutes.
The extremely high correlation between variables that is an inherent feature of near infrared (NI... more The extremely high correlation between variables that is an inherent feature of near infrared (NIR) spectroscopic data makes multivariate projection methods suitable as data analytical tools. Projection-based methods such as principal component analysis (PCA)1, partial least squares (PLS)2 regression and PLSdiscriminant analysis (PLS-DA)3 or soft independent modelling of class analogy (SIMCA)4 are therefore often used for data exploration, regression and classification respectively. In classical statistics, the canonical variates analysis (CVA) method was introduced in the 1930s by Fisher in order to optimise the group discriminative direction in multivariate space. CVA is an excellent method for estimating discriminative directions, but in contrast to e.g. PCA, the CVA method breaks down if the number of variables is larger than the number of samples and/or if the variables are highly correlated. In this respect, CVA is comparable to multiple linear regression (MLR), which suffers ...
Background/aims: Spectroscopy on human skin is a field that is being adopted increasingly because... more Background/aims: Spectroscopy on human skin is a field that is being adopted increasingly because of its rapidity and high reproducibility. Infrared reflectance (IR), near-infrared reflectance (NIR), and fluorescence spectroscopy have previously been applied to human skin in vivo to compare healthy and sick skin, including skin cancer, atopy, and leprosy. Exploratory data analysis/chemometrics is a tool for evaluating multivariate data such as spectroscopic measurements. The objective of this study was to explore the spectral variance spanned by people with normal integument, and to demonstrate the advantages of multivariate analysis to skin research. Methods: IR, NIR and fluorescence spectroscopy have been carried out in vivo on 216 volunteers' forearms before and after four tape strippings. The subjects were asked to fill in a questionnaire regarding factors suspected to influence the measurement results. Principal Component Analysis (PCA) was used to investigate whether the population can be divided into groups on the basis of their skin chemistry. Unless otherwise stated, the results are from the measurements prior to stripping. Results: In contrast to IR and fluorescence spectra, NIR spectra proved able to detect gender differences. By use of PCA, classifications on male and female subjects were observed from the IR and NIR measurements, and as an indication from the fluorescence measurements. The NIR and fluorescence measurements varied between elderly and young subjects. The largest variance in the fluorescence landscapes was seen between pigmented and non-pigmented skin. No connection was found between the spectroscopic measurements and smoking or drinking habits. Conclusions: Future spectroscopic skin investigations should be balanced as regards to gender and age, as these can possibly affect the measurement results. Chemometrics proved to be superior to traditional attempts of interpreting the spectra.
Dairy process monitoring by application of multivariate curve resolution using alternating least ... more Dairy process monitoring by application of multivariate curve resolution using alternating least squares is presented. Alternating least squares was used for resolving Fourier transform infrared spectral data from a dairy batch process in which lactose is enzymatically hydrolyzed to glucose and galactose. It was possible to extract four compounds (fat, lactose, and two other sugar components) from the spectral data obtained from nine process runs. Subsequently, the pure spectra obtained in this way were used to monitor the content of these compounds in two new process runs. In this way, alternating least squares made it possible to follow the hydrolysis process by Fourier transform infrared spectroscopy without the need for reference analyses. When the results were correlated to reference results for lactose, the accuracy was similar to that obtained when a partial least squares regression was performed on the same data; lactose correlation was 0.980 when alternating least squares was used and was 0.987 when partial least squares was used.
Uploads
Papers by Lars Nørgaard