| 1. |
- Abrahamsson, Christoffer, et al.
(författare)
-
Comparison of different variable selection methods conducted on NIR transmission measurements on intact tablets
- 2003
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - 0169-7439. ; 69:1-2, s. 3-12
-
Tidskriftsartikel (refereegranskat)abstract
- Near infrared (NIR) transmission spectroscopy is a promising method for fast quantitative measurements on pharmaceutical tablets, but there are still some problems to overcome in order to incorporate the technique as a control tool in tablet production. The main problem is the limited precision for multivariate calibrations based on NIR transmission data. The precision is affected by several factors, where one of the most important is which variable to include in the multivariate calibration model. In this work, four different methods for variable selection in partial least square (PLS) regression were studied and compared to a calibration made with manually selected wavelengths. The methods used were genetic algorithm (GA), iterative PLS (IPLS), uninformative variable elimination by PLS (UVE-PLS) and interactive variable selection for PLS (IVS-PLS). All methods improved the predictive abilities of the model compared to the model where the wavelengths were selected manually. For the data set used in this work, IVS-PLS and GA achieved the best results with improvements in prediction error by 20%, but further measurements and investigations have to be made before any general conclusion can be drawn.
|
|
| 2. |
- Kinser, J. M., et al.
(författare)
-
Multidimensional pulse image processing of chemical structure data
- 2000
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - 0169-7439 .- 1873-3239. ; 51:1, s. 115-124
-
Tidskriftsartikel (refereegranskat)abstract
- Recent advances in the understanding of the mammalian visual cortex have led to new approaches for image processing techniques. As a result of this, computer simulations using the proposed visual cortex model have become very useful in the field of image processing. Models of this kind have the ability to efficiently extract image segments, edges and texture. They operate by generating a set of pulse images (images with binary pixels) for a static input. These pulse images display synchronized activity of neighboring neurons, and it is these images in which the information about segments, edges and texture are displayed. Pulse image generation is dependent on autowaves that travel throughout the image. In order to extend pulse image processing to multidimensional data (i.e., data cubes), the autowaves are designed to expand in all of the cube's dimensions. In this fashion, pulse cubes can be created and the same analysis techniques that have been applied to two-dimensional pulse images can be applied to pulse image cubes. This paper examines and discusses multidimensional pulse image analysis applied to three-dimensional (3D) chemical structural data of 17 beta-estradiol.
|
|
| 3. |
- Verikas, Antanas, et al.
(författare)
-
Using artificial neural networks for process and system modelling
- 2003
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - Amsterdam : Elsevier. - 0169-7439 .- 1873-3239. ; 67:2, s. 187-191
-
Tidskriftsartikel (refereegranskat)abstract
- This letter concerns several papers, devoted to neural network-based process and system modelling, recently published in the Chemometrics and Intelligent Laboratory Systems journal. Artificial neural networks have proved themselves to be very useful in various modelling applications, because they can represent complex mapping functions and discover the representations using powerful learning algorithms. An optimal set of parameters for defining the functions is learned from examples by minimizing an error functional. In various practical applications, the number of examples available for estimating parameters of the models is rather limited. Moreover, to discover the best model, numerous candidate models must be trained and evaluated. In such thin-data situations, special precautions are to be taken to avoid erroneous conclusions. In this letter, we discuss three important issues, namely network initialization, over-fitting, and model selection, the right consideration of which can be of tremendous help in successful network design and can make neural modelling results more valuable.
|
|
| 4. |
- Antti, Henrik, et al.
(författare)
-
Statistical experimental design and partial least squares regression analysis of biofluid metabonomic NMR and clinical chemistry data for screening of adverse drug effects
- 2004
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - : Elsevier BV. - 0169-7439. ; 73:1, s. 139-49
-
Tidskriftsartikel (refereegranskat)abstract
- Metabonomic analysis is increasingly recognised as a powerful approach for delineating the integrated metabolic changes in biofluids and tissues due to toxicity, disease processes or genetic modification in whole animal systems. When dealing with complex biological data sets, as generated within metabonomics, as well as related fields such as genomics and proteomics, reliability and significance of identified biomarkers associated with specific states related to toxicity or disease are crucial in order to gain detailed and relevant interpretations of the metabolic fluxes in the studied systems. Since various physiological factors, such as diet, state of health, age, diurnal cycles, stress, genetic drift, and strain differences, affect the metabolic composition of biological matrices, it is of great importance to create statistically reliable decision tools for distinguishing between physiological and pathological responses in animal models. In the screening for new biomarkers or patterns of pathological dysfunction, methods providing statistically valid measures of effect-related changes will become increasingly important as the data within areas such as genomics, proteomics and metabonomics continues to grow in size and complexity. 1H NMR spectroscopy and mass spectrometry are the principal analytical platforms used to derive the data and, because extensively large data sets are required, as much consideration has to be given to optimum design of experiments (DoE) as for subsequent data analysis. Thus, statistical experimental design combined with partial least squares (PLS) regression is proposed as an efficient approach for undertaking metabonomic studies and for analysis of the results. The method was applied to data from a liver toxicology study in the rat using hydrazine as a model toxin. 1D projections of 2D J-resolved (J-RES) 1H NMR spectra and the corresponding clinical chemistry parameters of blood serum samples from control and dosed rats (30 and 90 mg/kg) collected at 48 and 168 h post dose were analysed. Confidence intervals for the PLS regression coefficients were used to create a statistical means for screening of biomarkers in the two combined data blocks (NMR and clinical chemistry data). PLS analysis was also used to reveal the correlation pattern between the two blocks of data as well as the within the two blocks according to dose, time and the interaction dose×time.
|
|
| 5. |
- Eriksson, Lennart, et al.
(författare)
-
Time-resolved QSAR: an approach to PLS modelling of three-way biological data
- 2004
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - : Elsevier BV. - 0169-7439. ; 73:1, s. 73-84
-
Tidskriftsartikel (refereegranskat)abstract
- This paper outlines a novel approach to the analysis of three-way Y-data in quantitative structure–activity relationship (QSAR) modelling. The new method represents a modification of an existing approach for multivariate modelling of batch process data. It is based on unfolding the three-way Y-matrix into a two-way matrix according to a sequential order of an external variable. In QSAR, time, pH, or temperature at which the biological data were gathered, are conceivably such external variables. Thus, unfolding can be done differently depending on the objective of the investigation, thereby shifting the focus of the QSAR analysis. The ensuing multivariate data analysis uses two levels of modelling. (1) On the lower (observation) level a projections to latent structures (PLS) model is developed between the unfolded biological data and the external variable. This model will identify compounds with biological data being sensitive to changes in the external variable (like time, pH, or temperature). (2) The scores of the lower level model are then re-arranged to enable the upper (QSAR) level model. In this model, a battery of structure descriptors (X) is related to the Y-matrix of scores of the lower level model. As an example, a series of 35 compounds and their anti-microbial activity towards the bacterial strain Escherichia coli CCM2260 is used. This biological activity has been determined at different times (2 to 10 h) and pH-values (pH 5.6 to 8.0).
|
|
| 6. |
- Gottfries, Johan, et al.
(författare)
-
Proteomics for drug target discovery
- 2004
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - : Elsevier B.V.. - 0169-7439. ; 73:1, s. 47-53
-
Tidskriftsartikel (refereegranskat)abstract
- Proteomics, genomics and metabonomics have, during the last decade, provided researchers with huge amounts of data. The choice of transformation of such data into useful information is dependent on the study aims and objectives. In the present study, projection methods (i.e., Principal Components Analysis [PCA] and Partial List Squares-Discriminant Analysis [PLS-DA]) were used to overview results from two-dimensional (2D) protein gel separations. The aim was to unravel possibilities for target discovery options via an in-depth understanding of quantified alterations in tissue or body fluid sample protein levels related to diseases. Two examples will be included comprising (1) data measured in cerebrospinal fluid (CSF) samples from diagnosed dementia patients and healthy volunteers, and (2) data from liver samples of drug-treated animals (i.e., Rosigltazone and Wy14643). The examples reveal clear clustering, using the protein levels as input, coinciding with the clinical diagnoses in example 1 and by treatment group in example 2.
|
|
| 7. |
- Olsson, Ing-Marie, et al.
(författare)
-
D-optimal onion designs in statistical molecular design
- 2004
-
Ingår i: Chemometrics and Intelligent Laboratory Systems. - : Elsevier BV. - 0169-7439. ; 73:1, s. 37-46
-
Tidskriftsartikel (refereegranskat)abstract
- Statistical molecular design (SMD) is a technique for selecting a representative (diverse) set of substances in combinatorial chemistry and QSAR, as well as other areas depending on optimising chemical structure. Two approaches often used in SMD are space filling (SF) and D-optimal (DO) designs.Space-filling designs provide good coverage of the physicochemical space but are not explicitly based on a model. For small design sizes, they perform similar to D-optimal designs, which maximize the determinant of the variance–covariance matrix. This leads to selection of the most extreme points of the candidate set and gives a minimal set of selected compounds with maximal diversity. However, the inner regions of the experimental domain are not well sampled by DO or small SF designs.We have developed and evaluated an approach to remedy the shortcomings of SF and DO designs in SMD. This new approach divides the candidate set into a number of subsets (“shells” or “layers”), and a D-optimal selection is made from each layer. This makes it possible to select representative sets of molecular structures throughout any property space, e.g., the physicochemical space, with reasonable design sizes. The number of selected molecules is easily controlled by varying (a) the number of layers and (b) the model on which the design is based.We outline here this new approach, the D-optimal onion design (DOOD). It is tested on two molecular data sets with varying size and compared with SF designs and ordinary DO designs. The designs have been evaluated with parameters, such as condition number, determinant, Tanimoto coefficients and Euclidean distances, as well as external evaluation of the resulting projection to latent structures (PLS) model.
|
|
| 8. |
|
|
| 9. |
|
|
| 10. |
|
|
