Bootstrapped Latin Partitions Research Articles

Near infrared spectroscopy combined with least squares support vector machines and fuzzy rule-building expert system applied to diagnosis of endometrial carcinoma

Objective: The feasibility of early diagnosis of endometrial carcinoma was studied by least squares support vector machines (LS-SVM) and fuzzy rule-building expert system (FuRES) that classified near infrared (NIR) spectra of tissues. Methods: NIR spectra of 77 specimens of endometrium were collected. The spectra were pretreated by principal component orthogonal signal correction (PC-OSC) and direct orthogonal signal correction (DOSC) methods to improve the signal-to-noise ratio (SNR) and remove the influences of background and baseline. The effects of modeling parameters were investigated using bootstrapped Latin-partition methods. Results: The optimal LS-SVM model of the PC-OSC pretreatment method successfully classified the samples with prediction accuracies of 96.8±1.4%. Conclusions: The proposed procedure proved to be rapid and convenient, which is suitable to be developed as a non-invasive diagnosis method for cancer tissue.

Baseline Correction Method Using an Orthogonal Basis for Gas Chromatography/Mass Spectrometry Data

A baseline correction method that uses basis set projection to estimate spectral backgrounds has been developed and applied to gas chromatography/mass spectrometry (GC/MS) data. An orthogonal basis was constructed using singular value decomposition (SVD) for each GC/MS two-way data object from a set of baseline mass spectra. A novel aspect of this baseline correction method is the regularization parameter that prevents overfitting that may produce negative peaks in the corrected mass spectra or ion chromatograms. The number of components in the basis, the regularization parameter, and the mass spectral range from which the spectra were sampled to construct the basis were optimized so that the projected difference resolution (PDR) or signal-to-noise ratio (SNR) was maximized. PDR is a metric similar to chromatographic resolution that indicates the separation of classes in a multivariate data space. This new baseline correction method was evaluated with two synthetic data sets and a real GC/MS data set. The prediction accuracies obtained by using the fuzzy rule-building expert system (FuRES) and partial least-squares-discriminant analysis (PLS-DA) as classifiers were compared and validated through bootstrapped Latin partition (BLP) between data before and after baseline correction. The results indicate that baseline correction of the two-way GC/MS data using the proposed methods resulted in a significant increase in average PDR values and prediction accuracies.

Classification of bacteria by simultaneous methylation–solid phase microextraction and gas chromatography/mass spectrometry analysis of fatty acid methyl esters

Direct methylation and solid-phase microextraction (SPME) were used as a sample preparation technique for classification of bacteria based on fatty acid methyl ester (FAME) profiles. Methanolic tetramethylammonium hydroxide was applied as a dual-function reagent to saponify and derivatize whole-cell bacterial fatty acids into FAMEs in one step, and SPME was used to extract the bacterial FAMEs from the headspace. Compared with traditional alkaline saponification and sample preparation using liquid-liquid extraction, the method presented in this work avoids using comparatively large amounts of inorganic and organic solvents and greatly decreases the sample preparation time as well. Characteristic gas chromatography/mass spectrometry (GC/MS) of FAME profiles was achieved for six bacterial species. The difference between Gram-positive and Gram-negative bacteria was clearly visualized with the application of principal component analysis of the GC/MS data of bacterial FAMEs. A cross-validation study using ten bootstrap Latin partitions and the fuzzy rule building expert system demonstrated 87 +/- 3% correct classification efficiency.

Automated Principal Component-Based Orthogonal Signal Correction Applied to Fused Near Infrared−Mid-Infrared Spectra of French Olive Oils

An approach for automating the determination of the number of components in orthogonal signal correction (OSC) has been devised. In addition, a novel principal component OSC (PC-OSC) is reported that builds softer models for removing background from signals and is much faster than the partial least-squares (PLS) based OSC algorithm. These signal correction methods were evaluated by classifying fused near- and mid-infrared spectra of French olive oils by geographic origin. Two classification methods, partial least-squares-discriminant analysis (PLS-DA) and a fuzzy rule-building expert system (FuRES), were used to evaluate the signal correction of the fused vibrational spectra from the olive oils. The number of components was determined by using bootstrap Latin partitions (BLPs) in the signal correction routine and maximizing the average projected difference resolution (PDR). The same approach was used to select the number of latent variables in the PLS-DA evaluation and perfect classification was obtained. Biased PLS-DA models were also evaluated that optimized the number of latent variables to yield the minimum prediction error. Fuzzy or soft classification systems benefit from background removal. The FuRES prediction results did not differ significantly from the results that were obtained using either the unbiased or biased PLS-DA methods, but was an order of magnitude faster in the evaluations when a sufficient number of PC-OSC components were selected. The importance of bootstrapping was demonstrated for the automated OSC and PC-OSC methods. In addition, the PLS-DA algorithms were also automated using BLPs and proved effective.

Application of Linear and Nonlinear Discrete Wavelet Transforms to MALDI-MS Measurements of Bacteria for Classification

The linear and nonlinear discrete wavelet transforms (DWTs) were used to compress matrix-assisted laser desorption/ionization mass spectra to address two key challenges: the relatively high noise level and the underdetermined format of the data set. By applying the DWT to MALDI-MS spectra, the spectra were simultaneously smoothed and compressed. Multivariate projected difference resolution was used to evaluate the effects of the linear and nonlinear DWT on classification. The cross-validation study using bootstrapped Latin partition and partial least-squares (PLS-2) has proved that the classification accuracy increased after data compression. The best result was obtained when using Fisher's criterion to choose wavelet coefficients for compression. With the aid of principal component analysis (PCA), different wavelet filters may provide different mathematical perspectives to visualize the clustering of bacteria. The effect of growth time was directly observed with wavelet transform, which could not be observed using the original spectra.

Forensic Application of Gas Chromatography–Differential Mobility Spectrometry with Two-Way Classification of Ignitable Liquids from Fire Debris

With respect to the emerging role of forensic science for arson investigation, a low cost and promising onsite detection method for ignitable liquids is desirable. Gas chromatography-differential mobility spectrometry (GC-DMS) was investigated as a tool for analysis of ignitable liquids from fire debris. Headspace solid-phase microextraction (SPME) was applied as the preconcentration and sampling method. The combined information afforded by gas chromatography and differential mobility spectrometry provided unique two-way patterns for each sample of ignitable liquid. Two-way GC-DMS data were classified into one of seven ignitable liquids using a fuzzy rule-building expert system (FuRES). The performance of the classifier was validated using bootstrap Latin partitions (BLPs) and also compared to optimized partial least-squares (PLS) classifiers. Better prediction results can be obtained by using two-way GC-DMS data than only using one-way total ion chromatograms or integrated differential mobility spectra. FuRES models constructed with the neat ignitable liquids identified the spiked samples from simulated fire debris with 99.07 +/- 0.04% accuracy.

Bootstrap classification and point-based feature selection from age-staged mouse cerebellum tissues of matrix assisted laser desorption/ionization mass spectra using a fuzzy rule-building expert system

A bootstrap method for point-based detection of candidate biomarker peaks has been developed from pattern classifiers. Point-based detection methods are advantageous in comparison to peak-based methods. Peak determination and selection are problematic when spectral peaks are not baseline resolved or on a varying baseline. The benefit of point-based detection is that peaks can be globally determined from the characteristic features of the entire data set (i.e., subsets of candidate points) as opposed to the traditional method of selecting peaks from individual spectra and then combining the peak list into a data set. The point-based method is demonstrated to be more effective and efficient using a synthetic data set when compared to using Mahalanobis distance for feature selection. In addition, probabilities that characterize the uniqueness of the peaks are determined. This method was applied for detecting peaks that characterize age-specific patterns of protein expression of developing and adult mouse cerebella from matrix assisted laser desorption/ionization (MALDI) mass spectrometry (MS) data. The mice comprised three age groups: 42 adults, 19 14-day-old pups, and 16 7-day-old pups. Three sequential spectra were obtained from each tissue section to yield 126, 57 and 48 spectra for adult, 14-day-old pup, and 7-day-old pup spectra, respectively. Each spectrum comprised 71,879 mass measurements in a range of 3.5–50 kDa. A previous study revealed that 846 unique peaks were detected that were consistent for 50% of the mice in each age group (C. Laurent, D.F. Levinson, S.A. Schwartz, P.B. Harrington, S.P. Markey, R.M. Caprioli, P. Levitt, Direct profiling of the cerebellum by MALDI MS: a methodological study in postnatal and adult mouse, J. Neurosci. Res. 81 (2005) 613–621.). A fuzzy rule-building expert system (FuRES) was applied to investigate the correlation of age with features in the MS data. FuRES detected two outlier pup-14 spectra. Prediction was evaluated using 100 bootstrap samples of 2 Latin-partitions (i.e., 50:50 split between training and prediction set) of the mice. The spectra without the outliers yielded classification rates of 99.1 ± 0.1%, 90.1 ± 0.8%, and 97.0 ± 0.6% for adults, 14-day-old pups, and 7-day-old pups, respectively. At a 95% level of significance, 100 bootstrap samples disclosed 35 adult and 21 pup distinguishing peaks for separating adults from pups; and 8 14-day-old and 15 7-day-old predictive peaks for separating 14-day-old pup from 7-day-old pup spectra. A compressed matrix comprising 40,393 points that were outside the 95% confidence intervals of one of the two FuRES discriminants was evaluated and the classification improved significantly for all classes. When peaks that satisfied a quality criterion were integrated, the 55 integrated peak areas furnished significantly improved classification for all classes: the selected peak areas furnished classification rates of 100%, 97.3 ± 0.6%, and 97.4 ± 0.3% for adult, 14-day-old pups, and 7-day-old pups using 100 bootstrap Latin partitions evaluations with the predictions averaged. When the bootstrap size was increased to 1000 samples, the results were not significantly affected. The FuRES predictions were consistent with those obtained by discriminant partial least squares (DPLS) classifications.

Statistical validation of classification and calibration models using bootstrapped Latin partitions

Unbiased evaluation of classification and calibration methods is important, especially as these methods are applied to increasingly complex data sets that are under-determined. Precision bounds, such as confidence intervals, are required for interpreting any experimental result. Using bootstrapped Latin partitions to evaluate classification and calibration models, bounds on the average predictions were obtained. These bounds characterize sources of variation attributed to building the model and the composition of the training set with respect to the test set. Furthermore, precision bounds on the average of the model-variable loadings allow the significance of characteristic features to be estimated. The procedure for bootstrapped Latin partitions is given and demonstrated with synthetic data sets for classification using linear discriminant analysis and fuzzy rule-building expert systems, and for calibration using partial least squares regression with one and three properties. All analyses were implemented on a personal computer with the longest evaluation requiring 6-h processing time. Analysis of variance and matched sample t-tests were also used to demonstrate the statistical power of these tests.