A brief note on application of domain-invariant PLS for adapting near-infrared spectroscopy calibrations between different physical forms of samples.

Near-infrared (NIR) calibration models are widely developed and routinely used for the prediction of physicochemical properties of samples. However, the main challenge with NIR models is that they are highly specific to the physical form of the samples. For example, a NIR calibration established for solid samples can usually not be used for the same samples in powdered form. Domain adaption (DA) techniques, such as domain invariant partial least-squares (di-PLS) regression, have recently appeared in the chemometric domain which allow adapting NIR calibrations for new sample-/instrument- or environment-associated conditions in a standard free manner. A practical use case of di-PLS can be assumed as the adaption of NIR calibration models to be used in different physical forms of samples. In this contribution we show, for the first time, application of di-PLS regression analysis for adapting a near-infrared (NIR) calibration for solid rice kernels to be used on powdered rice flour without the need for new reference measurements for the latter. di-PLS is a domain adaption technique that removes the differences between different but related data sources (i.e. domains) to reach generalized models. The study found that di-PLS allowed a direct adaption of calibration based on solid rice kernels to be used on powdered rice flour without requiring any reference protein measurements for the latter. Our results suggest that DA tools, such as di-PLS, can support a wider usage of chemometric calibrations especially when models need to be adapted to different physical forms of the same samples.

Apparent penetration depth in attenuated total reflection Fourier-transform infrared (ATR-FTIR) spectroscopy of Allium cepa L. epidermis and cuticle.

Over the past decades, ATR-FTIR has emerged as promising tool for the identification of plants at the genus and (sub-) species level through surface measurements of intact leaves. Theoretical considerations regarding the penetration depth of the evanescent wave into the sample and the thickness of plant leaf cuticles suggest that the structure and composition of the cuticle represent universal taxonomic markers. However, experimental evidence for this hypothesis is scarce. In the current contribution, we present results of a series of simple experiments on epidermal monolayers derived from the bulbs of Allium cepa L. (Amaryllidaceae) as a model system to study the effect of an IR active probe located beyond the theoretical penetration depth of the evanescent wave. We found that this probe had a significant influence on the ATR-FTIR spectra for up to 4 epidermal layers stacked on top of each other corresponding to a total thickness of around 60 µm, exceeding the theoretical penetration depth of the evanescent wave by a factor of around 20. Altogether, our data indicate a major discrepancy between theory and practice in ATR-FTIR spectroscopy in general and provide strong evidence that in general plant leaf spectra cannot be fully explained by the structure and composition of the cuticle alone.

Probeless non-invasive near-infrared spectroscopic bioprocess monitoring using microspectrometer technology.

Real-time measurements and adjustments of critical process parameters are essential for the precise control of fermentation processes and thus for increasing both quality and yield of the desired product. However, the measurement of some crucial process parameters such as biomass, product, and product precursor concentrations usually requires time-consuming offline laboratory analysis. In this work, we demonstrate the in-line monitoring of biomass, penicillin (PEN), and phenoxyacetic acid (POX) in a Penicilliumchrysogenum fed-batch fermentation process using low-cost microspectrometer technology operating in the near-infrared (NIR). In particular, NIR reflection spectra were taken directly through the glass wall of the bioreactor, which eliminates the need for an expensive NIR immersion probe. Furthermore, the risk of contaminations in the reactor is significantly reduced, as no direct contact with the investigated medium is required. NIR spectra were acquired using two sensor modules covering the spectral ranges 1350-1650 nm and 1550-1950 nm. Based on offline reference analytics, partial least squares (PLS) regression models were established for biomass, PEN, and POX either using data from both sensors separately or jointly. The established PLS models were tested on an independent validation fed-batch experiment. Root mean squared errors of prediction (RMSEP) were 1.61 g/L, 1.66 g/L, and 0.67 g/L for biomass, PEN, and POX, respectively, which can be considered an acceptable accuracy comparable with previously published results using standard process spectrometers with immersion probes. Altogether, the presented results underpin the potential of low-cost microspectrometer technology in real-time bioprocess monitoring applications. Graphical abstract.

Broadband near-infrared hyperspectral single pixel imaging for chemical characterization.

We introduce a compressive sensing based approach for single pixel hyperspectral chemical imaging in a broad spectral range in the near-infrared. Fully integrated MEMS based Fabry-Pérot tunable filter spectrometers and a digital micro-mirror device were employed to achieve spectral and spatial resolution, respectively. The available spectral range from 1500 to 2200 nm covers molecular overtone vibrations enabling chemical identification. Hyperspectral images of different adhesives deposited on a textile were recorded revealing their chemical composition. Furthermore, spectrally resolved near-infrared images with compression rates up to 90% are presented. The approach of single pixel imaging illustrates a promising technology for the infrared spectral range superior to conventionally used costly focal plane arrays.

Domain-Invariant Partial-Least-Squares Regression.

Multivariate calibration models often fail to extrapolate beyond the calibration samples because of changes associated with the instrumental response, environmental condition, or sample matrix. Most of the current methods used to adapt a source calibration model to a target domain exclusively apply to calibration transfer between similar analytical devices, while generic methods for calibration-model adaptation are largely missing. To fill this gap, we here introduce domain-invariant partial-least-squares (di-PLS) regression, which extends ordinary PLS by a domain regularizer in order to align the source and target distributions in the latent-variable space. We show that a domain-invariant weight vector can be derived in closed form, which allows the integration of (partially) labeled data from the source and target domains as well as entirely unlabeled data from the latter. We test our approach on a simulated data set where the aim is to desensitize a source calibration model to an unknown interfering agent in the target domain (i.e., unsupervised model adaptation). In addition, we demonstrate unsupervised, semisupervised, and supervised model adaptation by di-PLS on two real-world near-infrared (NIR) spectroscopic data sets.

Calibration model maintenance in melamine resin production: Integrating drift detection, smart sample selection and model adaptation.

The physico-chemical properties of Melamine Formaldehyde (MF) based thermosets are largely influenced by the degree of polymerization (DP) in the underlying resin. On-line supervision of the turbidity point by means of vibrational spectroscopy has recently emerged as a promising technique to monitor the DP of MF resins. However, spectroscopic determination of the DP relies on chemometric models, which are usually sensitive to drifts caused by instrumental and/or sample-associated changes occurring over time. In order to detect the time point when drifts start causing prediction bias, we here explore a universal drift detector based on a faded version of the Page-Hinkley (PH) statistic, which we test in three data streams from an industrial MF resin production process. We employ committee disagreement (CD), computed as the variance of model predictions from an ensemble of partial least squares (PLS) models, as a measure for sample-wise prediction uncertainty and use the PH statistic to detect changes in this quantity. We further explore supervised and unsupervised strategies for (semi-)automatic model adaptation upon detection of a drift. For the former, manual reference measurements are requested whenever statistical thresholds on Hotelling's T2 and/or Q-Residuals are violated. Models are subsequently re-calibrated using weighted partial least squares in order to increase the influence of newer samples, which increases the flexibility when adapting to new (drifted) states. Unsupervised model adaptation is carried out exploiting the dual antecedent-consequent structure of a recently developed fuzzy systems variant of PLS termed FLEXFIS-PLS. In particular, antecedent parts are updated while maintaining the internal structure of the local linear predictors (i.e. the consequents). We found improved drift detection capability of the CD compared to Hotelling's T2 and Q-Residuals when used in combination with the proposed PH test. Furthermore, we found that active selection of samples by active learning (AL) used for subsequent model adaptation is advantageous compared to passive (random) selection in case that a drift leads to persistent prediction bias allowing more rapid adaptation at lower reference measurement rates. Fully unsupervised adaptation using FLEXFIS-PLS could improve predictive accuracy significantly for light drifts but was not able to fully compensate for prediction bias in case of significant lack of fit w.r.t. the latent variable space.

Quality Control of Valerianae Radix by Attenuated Total Reflection Fourier Transform Infrared (ATR-FTIR) Spectroscopy.

(Acetoxy-)valerenic acid and total essential oil content are important quality attributes of pharmacy grade valerian root (Valerianae radix). Traditional analysis of these quantities is time-consuming and necessitates (harmful) solvents. Here we investigated an application of attenuated total reflection Fourier transform infrared spectroscopy for extractionless analysis of these quality attributes on a representative sample comprising 260 wild-crafted individuals covering the Central European taxonomic diversity of the Valeriana officinalis L. s. l. species aggregate with its three major ploidy cytotypes (i.e., di-, tetra- and octoploid). Calibration models were built by orthogonal partial least squares regression for quantitative analysis of (acetoxy-)valerenic acid and total essential oil content. For the latter, we propose a simplistic protocol involving apolar extraction followed by gas chromatography as a reference method for multivariate calibration in order to handle the analysis of samples taken from individual plants. We found good predictive ability of chemometric models for quantification of valerenic acid, acetoxyvalerenic acid, total sesquiterpenoid acid, and essential oil content with a root mean squared error of cross-validation of 0.064, 0.043, and 0.09 and root mean squared error of prediction of 0.066, 0.057, and 0.09 (% content), respectively. Orthogonal partial least squares discriminant analysis revealed good discriminability between the most productive phenotype (i.e., the octoploid cytotype) in terms of sesquiterpenoid acids, and the less productive ones (i.e., di- and tetraploid). All in all, our results demonstrate the application of attenuated total reflection Fourier transform infrared spectroscopy for rapid, extractionless estimation of the most important quality attributes of valerian root and minimally invasive identification of the most productive phenotype in terms of sesquiterpenoid acids.

Automatic feed phase identification in multivariate bioprocess profiles by sequential binary classification.

In this paper, we propose a new strategy for retrospective identification of feed phases from online sensor-data enriched feed profiles of an Escherichia Coli (E. coli) fed-batch fermentation process. In contrast to conventional (static), data-driven multi-class machine learning (ML), we exploit process knowledge in order to constrain our classification system yielding more parsimonious models compared to static ML approaches. In particular, we enforce unidirectionality on a set of binary, multivariate classifiers trained to discriminate between adjacent feed phases by linking the classifiers through a one-way switch. The switch is activated when the actual classifier output changes. As a consequence, the next binary classifier in the classifier chain is used for the discrimination between the next feed phase pair etc. We allow activation of the switch only after a predefined number of consecutive predictions of a transition event in order to prevent premature activation of the switch and undertake a sensitivity analysis regarding the optimal choice of the (time) lag parameter. From a complexity/parsimony perspective the benefit of our approach is three-fold: i) The multi-class learning task is broken down into binary subproblems which usually have simpler decision surfaces and tend to be less susceptible to the class-imbalance problem. ii) We exploit the fact that the process follows a rigid feed cycle structure (i.e. batch-feed-batch-feed) which allows us to focus on the subproblems involving phase transitions as they occur during the process while discarding off-transition classifiers and iii) only one binary classifier is active at the time which keeps effective model complexity low. We further use a combination of logistic regression and Lasso (i.e. regularized logistic regression, RLR) as a wrapper to extract the most relevant features for individual subproblems from the whole set of high-dimensional sensor data. We train different soft computing classifiers, including decision trees (DT), k-nearest neighbors (k-NN), support vector machines (SVM) and an own developed fuzzy classifier and compare our method with conventional multi-class ML. Our results show a remarkable out-performance of the here proposed method over static ML approaches in terms of accuracy and robustness. We achieved close to error free feed phase classification while reducing the misclassification rates in 17 out of 20 investigated test cases in the range between 39% and 98.2% depending on feature set and classifier architecture. Models trained on features based on selection by RLR significantly outperformed those trained on features suggested by experts and their predictive performance was considerably less affected by the choice of the lag parameter.