A multivariate approach for the statistical evaluation of near-infrared chemical images using Symmetry Parameter Image Analysis (SPIA)

Near-Infrared Chemical Imaging (NIR-CI) is rapidly gaining importance for the analysis of complex intermediate and final drug products. The availability of both spectral information from the sample and spatial information on the distribution of individual components offers access to greater understanding of manufacturing processes in many stages of pharmaceutical production. One major aspect in terms of chemical imaging is data analysis, since each measurement (image) generates a data cube containing several thousands of spectra (i.e., one spectrum per image pixel). The visual interpretation of component distribution (e.g., homogeneity) is an important issue but subjective. Chemometric methods are therefore required to extract qualitative and quantitative information from each image and enable comparison of several images. In this work, we describe a novel approach for the statistical evaluation of NIR-CI in terms of a multivariate treatment of univariate statistical descriptors characterizing image pixel (e.g., skewness and kurtosis). This technique was called by the authors "Symmetry Parameter Image Analysis" (SPIA), since it enables assessing the symmetry of pixel distributions in terms of different sample attributes. That approach is an innovative way of reporting results with a straightforward relation with attributes such as homogeneity, thus providing the basis for setting up acceptance criteria for good processing conditions or sample homogeneity. Furthermore, this procedure is applicable to determine product variability for large data sets without the need for explicit consideration of each image as its main attributes have been captured by the pixel distributions and their univariate descriptors. The approach is described by means of data obtained by NIR-CI on a powder blend case study (process application). Additionally, SPIA was used for the qualitative classification of tablets (sample application), showing that the approach can be generalized to set up criteria for sample-to-sample similarity and be useful in establishing criteria for e.g., counterfeiting.

Near-infrared chemical imaging (NIR-CI) for counterfeit drug identification--a four-stage concept with a novel approach of data processing (Linear Image Signature)

A new stage concept was developed to reliably identify counterfeit tablets which are very similar to the genuine drug product. This concept combines single-point near-infrared spectroscopy (NIRS) and near-infrared chemical imaging (NIR-CI) with statistical variance analysis. The advantage of NIR-CI over NIRS is the potential to determine not only the amount, but also the spatial distribution of ingredients within a single tablet. Previously published NIR-CI studies used homogeneity as a key indicator for the identification of counterfeits. The state of the art approach for estimating homogeneity is to record the average and % standard deviation of predicted classification scores (i.e. concentrations) for a given component within a specimen. A disadvantage of this approach is the partial loss of spatial information. In view of this, we developed a new method using much more of the spatial information for the estimation of homogeneity. The method is based on (1) summation and unfolding of multidimensional predicted classification scores, which results in a Linear Image Signature (LIS) and (2) multivariate LIS data analysis (LIS-MVA). It could be demonstrated that this kind of NIR-CI data analysis represents an innovative approach for the identification of counterfeit tablets. Moreover, this procedure is applicable to determine the product variability, i.e. process signature of a given product thus being a valuable tool within the Quality by Design (QbD) approach of the ICH Q8 guideline.

Development of PLS calibration models from UV-Vis spectra for TOC estimation at the outlet of a fuel park wastewater treatment plant

In the present work ultraviolet (UV)-visible spectra of water samples collected at the outlet of a fuel park wastewater treatment plant, including biological treatment, were acquired and used for the development of partial least squares (PLS) calibration models for the fast and simple estimation of total organic carbon (TOC). Three different PLS models were developed and compared on the basis of a common spectral range. The first model was obtained using spectra of raw samples, the second using spectra of diluted samples, to assess signal saturation in the UV region, and the third using spectra of both diluted and raw samples, in order to expand the narrow interval of TOC concentration values present in the original dataset. The root mean squared error of cross-validation values for the developed PLS models were 2.3, 1.0 and 4.4 mg Cl(-1), respectively, and the validation results where highly satisfactory (root mean squared error of prediction values of 1.8, 0.8 and 4.5 mg Cl(-1), respectively).

Dealing with gene expression missing data

Compared evaluation of different methods is presented for estimating missing values in microarray data: weighted K-nearest neighbours imputation (KNNimpute), regression-based methods such as local least squares imputation (LLSimpute) and partial least squares imputation (PLSimpute) and Bayesian principal component analysis (BPCA). The influence in prediction accuracy of some factors, such as methods' parameters, type of data relationships used in the estimation process (i.e. row-wise, column-wise or both), missing rate and pattern and type of experiment [time series (TS), non-time series (NTS) or mixed (MIX) experiments] is elucidated. Improvements based on the iterative use of data (iterative LLS and PLS imputation--ILLSimpute and IPLSimpute), the need to perform initial imputations (modified PLS and Helland PLS imputation--MPLSimpute and HPLSimpute) and the type of relationships employed (KNNarray, LLSarray, HPLSarray and alternating PLS--APLSimpute) are proposed. Overall, it is shown that data set properties (type of experiment, missing rate and pattern) affect the data similarity structure, therefore influencing the methods' performance. LLSimpute and ILLSimpute are preferable in the presence of data with a stronger similarity structure (TS and MIX experiments), whereas PLS-based methods (MPLSimpute, IPLSimpute and APLSimpute) are preferable when estimating NTS missing data.

UV spectra analysis for water quality monitoring in a fuel park wastewater treatment plant

In the context of the high application potentials for on-line measurements in wastewater quality monitoring, UV spectroscopy has received recent attention. In the present work UV spectrophotometric analyses were coupled to principal component analysis (PCA) and cluster analysis (CA) to characterize samples taken from a fuel park wastewater treatment plant and to attempt preliminary contaminant identification in the treated wastewater. The score plot resulting from PCA identified two different groups of spectra, one including the influents to the biological reactor and the other the treated wastewater samples. Among the latter, weekday and weekend samples could be further distinguished. The same groups of samples were identified in a dendrogram from CA. The score plot and the dendrogram also allowed the tentative identification of employed process chemicals (lubricant and detergents) as residual contaminants in the treated effluent.

Capability index--a statistical process control tool to aid in udder health control in dairy herds

Bulk milk somatic cell count (BMSCC) averages have been used to evaluate udder health both at the individual or the herd level as well as milk quality and hygiene. The authors show that the BMSCC average is not the best tool to be used in udder health control programs and that it can be replaced with advantage by the capability index (Cpk). The Cpk is a statistical process control tool traditionally used by engineers to validate, monitor, and predict the expected behavior of processes or machines. The BMSCC data of 13 consecutive months of production from 414 dairy herds as well as SCC from all cows in the DHI program from 264 herds in the same period were collected. The Cpk and the annual BMSCC average (AAVG) of all the herds were calculated. Confronting the herd's performance explained by the Cpk and AAVG with the European Union (EU) official limit for BMSCC of 400,000 cells/mL, it was noticed that the Cpk accurately classified the compliance of the 414 farms, whereas the AAVG misclassified 166 (40%) of the 414 selected farms. The annual prevalence of subclinical mastitis (SMP) of each herd was calculated with individual SCC data from the same 13-mo period. Cows with more than 200,000 SCC/mL were considered as having subclinical mastitis. A logistic regression model to relate the Cpk and the herd's subclinical mastitis prevalence was calculated. The model is: SMPe = 0.475 e(-0.5286 x Cpk). The validation of the model was carried out evaluating the relation between the observed SMP and the predicted SMPe, in terms of the linear correlation coefficient (R2) and the mean difference between SMP and SMPe (i.e., mean square error of prediction). The validation suggests that our model can be used to estimate the herd's SMP with the herd's Cpk. The Cpk equation relates the herd's BMSCC with the EU official SCC limit, thus the logistic regression model enables the adoption of critical limits for subclinical mastitis, taking into consideration the legal standard for SCC.

Morphology and viability analysis of Streptomyces clavuligerus in industrial cultivation systems

The effects of varying inoculum age and production scale upon the morphology and viability of Streptomyces clavuligerus were studied by analyzing visible and fluorescent light images acquired throughout pilot-plant and pre-industrial scale fermentations. Changes in production scale reveal that in 5 m(3) fermentors, the maximum hyphal area obtained is double the value obtained in 0.5 m(3) fermentors. It is probably due to the higher shear stresses acting upon hyphae in the 0.5 m(3) fermentor caused by higher tip speeds observed in these. The morphological quantification based on elongation and branching rates allowed fermentations to be pattern classified into distinct physiological time zones namely elongation, branching, fragmentation, etc. The general pattern observed for fermentations inoculated with late exponential phase inocula was similar to the pattern of fermentations run with stationary phase inocula except that both the elongation and branching periods started earlier in the former case. Using the available staining technique and image acquisition system, the viability seemed to be generally high and constant throughout the time course of all the studied fermentations.

Multiblock PLS analysis of an industrial pharmaceutical process

The performance of an industrial pharmaceutical process (production of an active pharmaceutical ingredient by fermentation, API) was modeled by multiblock partial least squares (MBPLS). The most important process stages are inoculum production and API production fermentation. Thirty batches (runs) were produced according to an experimental planning. Rather than merging all these data into a single block of independent variables (as in ordinary PLS), four data blocks were used separately (manipulated and quality variables for each process stage). With the multiblock approach it was possible to calculate weights and scores for each independent block. It was found that the inoculum quality variables were highly correlated with API production for nominal fermentations. For the nonnominal fermentations, the manipulations of the fermentation stage explained the amount of API obtained (especially the pH and biomass concentration). Based on the above process analysis it was possible to select a smaller set of variables with which a new model was built. The amount of variance predicted of the final API concentration (cross-validation) for this model was 82.4%. The advantage of the multiblock model over the standard PLS model is that the contributions of the two main process stages to the API volumetric productivity were determined.

Real time monitoring biomass concentration in Streptomyces clavuligerus cultivations with industrial media using a capacitance probe

On-line monitoring biomass concentration in mycelial fed-batch cultivations of Streptomyces clavuligerus grown with soluble and partially insoluble complex media, was investigated with an in-situ capacitance probe fitted to an industrial pilot-plant tank. Standard off-line and on-line biomass determinations, including cell dry weight, packed mycelial volume, viscosity, DNA concentration and total CO(2) evolution in the exhaust gases, were performed throughout the experiments and compared to on-line capacitance measurements. Linear relations between capacitance and all other measurements were developed for both media that hold only in defined process phases, depending on the biomass state and the amount of insoluble matter present. For the industrial complex culture media good linear relations were obtained in the fast growth phase between capacitance and DNA concentration and total CO(2) evolution, while in the subsequent transition and stationary phases only with apparent viscosity was a reasonable correlation found. The capacitance probe was shown to be a valuable tool for real-time monitoring biomass concentration in industrial-like cultivation of mycelial streptomycetes.

Effects of preculture variability on clavulanic acid fermentation

The production profile of clavulanic acid by Streptomyces clavuligerus was shown to be strongly dependent on inoculum activity. Two sets of fermentations (A and B) were investigated at industrial pilot-plant scale using complex media. Type A fermentations were inoculated using late exponential growth phase mycelia. Type B fermentations were inoculated using mycelia harvested at stationary phase. Productivities throughout type A fermentations were consistently higher than type B, reaching a maximum at about 70 h and then decaying to the same final productivities at 140 h of type B runs. Several scheduling alternatives, based on combinations of the two inocula types and different fermentation lengths, were compared in terms of the overall process economics (fermentation and downstream). An increase of ca. 22% on the overall process profit is predicted using late exponential growth phase inocula and a fermentation duration of only 96 h. A new operating strategy was thus proposed for inoculum production based on the control of preculture activity using off-gas analysis. This method ensures higher productivity and better batch-to-batch reproducibility of clavulanic acid fermentations than traditional methods based on constant age inocula.

Mathematical modelling of industrial pilot-plant penicillin-G fed-batch fermentations

Penicillin-G fermentation with industrial media in 1 m3 stirred tank bioreactors was studied. A model based on the Bajpai-Reuss model structure was developed. Under typical production conditions catabolite repression is nonidentifiable and extensive mycelium differentiation occurs. Thus, the original model was reformulated, neglecting glucose repression of penicillin production and including biomass autolysis. The multi-substrate nature of industrial media was critically analysed. By combining the two most important carbon substrates present, a simple and applicable model was obtained. Model predictions agreed well with experimental data and reproduced the general characteristics observed in the fermentations. The predictive power of the model was tested for fermentations with different sugar feed rate profiles and raw materials (corn-steep liquor and sugar syrup). Several aspects of parameter estimation and model development are discussed on the basis of direct experimental data inspection and a sensitivity analysis of model parameters.