Predicting subacute ruminal acidosis from milk mid-infrared estimated fatty acids and machine learning on Canadian commercial dairy herds.

Our objective was to validate the possibility of detecting SARA from milk Fourier transform mid-infrared spectroscopy estimated fatty acids (FA) and machine learning. Subacute ruminal acidosis is a common condition in modern commercial dairy herds for which the diagnostic remains challenging due to its symptoms often being subtle, nonexclusive, and not immediately apparent. This observational study aimed at evaluating the possibility of predicting SARA by developing machine learning models to be applied to farm data and to provide an estimated portrait of SARA prevalence in commercial dairy herds. A first data set composed of 488 milk samples of 67 cows (initial DIM = 8.5 ± 6.18; mean ± SD) from 7 commercial dairy farms and their corresponding SARA classification (SARA+ if rumen pH <6.0 for 300 min, else SARA-) was used for the development of machine learning models. Three sets of predictive variables: i) milk major components (MMC), ii) milk FA (MFA), and iii) MMC combined with MFA (MMCFA) were submitted to 3 different algorithms, namely Elastic net (EN), Extreme gradient boosting (XGB), and Partial least squares (PLS), and evaluated using 3 different scenarios of cross-validation. Accuracy, sensitivity, and specificity of the resulting 27 models were analyzed using a linear mixed model. Model performance was not significantly affected by the choice of algorithm. Model performance was improved by including fatty acids estimations (MFA and MMCFA as opposed to MMC alone). Based on these results, one model was selected (algorithm: EN; predictive variables: MMCFA; 60.4, 65.4, and 55.3% of accuracy, sensitivity, and specificity, respectively) and applied to a large data set comprising the first test-day record (milk major components and FA within the first 70 DIM of 211,972 Holstein cows (219,503 samples) collected from 3001 commercial dairy herds. Based on this analysis, the within-herd SARA prevalence of commercial farms was estimated at 6.6 ± 5.29% ranging from 0 to 38.3%. A subsequent linear mixed model was built to investigate the herd-level factors associated to higher within-herd SARA prevalence. Milking system, proportion of primiparous cows, herd size and seasons were all herd-level factors affecting SARA prevalence. Furthermore, milk production was positively, and milk fat yield negatively associated with SARA prevalence. Due to their moderate levels of accuracy, the SARA prediction models developed in our study, using data from continuous pH measurements on commercial farms, are not suitable for diagnostic purpose. However, these models can provide valuable information at the herd level.

Comparison of the genetic characteristics of directly measured and Fourier-transform mid-infrared-predicted bovine milk fatty acids and proteins.

Fourier-transform mid-infrared (FT-MIR) spectroscopy is a high-throughput and inexpensive methodology used to evaluate concentrations of fat and protein in dairy cattle milk samples. The objective of this study was to compare the genetic characteristics of FT-MIR predicted fatty acids and individual milk proteins with those that had been measured directly using gas and liquid chromatography methods. The data used in this study was based on 2,005 milk samples collected from 706 Holstein-Friesian × Jersey animals that were managed in a seasonal, pasture-based dairy system, with milk samples collected across 2 consecutive seasons. Concentrations of fatty acids and protein fractions in milk samples were directly determined by gas chromatography and high-performance liquid chromatography, respectively. Models to predict each directly measured trait based on FT-MIR spectra were developed using partial least squares regression, with spectra from a random selection of half the cows used to train the models, and predictions for the remaining cows used as validation. Variance parameters for each trait and genetic correlations for each pair of measured/predicted traits were estimated from pedigree-based bivariate models using REML procedures. A genome-wide association study was undertaken using imputed whole-genome sequence, and quantitative trait loci (QTL) from directly measured traits were compared with QTL from the corresponding FT-MIR predicted traits. Cross-validation prediction accuracies based on partial least squares for individual and grouped fatty acids ranged from 0.18 to 0.65. Trait prediction accuracies in cross-validation for protein fractions were 0.53, 0.19, and 0.48 for α-casein, ß-casein, and κ-casein, 0.31 for α-lactalbumin, 0.68 for ß-lactoglobulin, and 0.36 for lactoferrin. Heritability estimates for directly measured traits ranged from 0.07 to 0.55 for fatty acids; and from 0.14 to 0.63 for individual milk proteins. For FT-MIR predicted traits, heritability estimates were mostly higher than for the corresponding measured traits, ranging from 0.14 to 0.46 for fatty acids, and from 0.30 to 0.70 for individual proteins. Genetic correlations between directly measured and FT-MIR predicted protein fractions were consistently above 0.75, with the exceptions of C18:0 and C18:3 cis-3, which had genetic correlations of 0.72 and 0.74, respectively. The GWAS identified trait QTL for fatty acids with likely candidates in the DGAT1, CCDC57, SCD, and GPAT4 genes. Notably, QTL for SCD were largely absent in the FT-MIR predicted traits, and QTL for GPAT4 were absent in directly measured traits. Similarly, for directly measured individual proteins, we identified QTL with likely candidates in the CSN1S1, CSN3, PAEP, and LTF genes, but the QTL for CSN3 and LTF were absent in the FT-MIR predicted traits. Our study indicates that genetic correlations between directly measured and FT-MIR predicted fatty acid and protein fractions are typically high, but that phenotypic variation in these traits may be underpinned by differing genetic architecture.

Predicting cow milk quality traits from routinely available milk spectra using statistical machine learning methods.

Numerous statistical machine learning methods suitable for application to highly correlated features, as those that exist for spectral data, could potentially improve prediction performance over the commonly used partial least squares approach. Milk samples from 622 individual cows with known detailed protein composition and technological trait data accompanied by mid-infrared spectra were available to assess the predictive ability of different regression and classification algorithms. The regression-based approaches were partial least squares regression (PLSR), ridge regression (RR), least absolute shrinkage and selection operator (LASSO), elastic net, principal component regression, projection pursuit regression, spike and slab regression, random forests, boosting decision trees, neural networks (NN), and a post-hoc approach of model averaging (MA). Several classification methods (i.e., partial least squares discriminant analysis (PLSDA), random forests, boosting decision trees, and support vector machines (SVM)) were also used after stratifying the traits of interest into categories. In the regression analyses, MA was the best prediction method for 6 of the 14 traits investigated [curd firmness at 60 min, αS1-casein (CN), αS2-CN, κ-CN, α-lactalbumin, and ß-lactoglobulin B], whereas NN and RR were the best algorithms for 3 traits each (rennet coagulation time, curd-firming time, and heat stability, and curd firmness at 30 min, ß-CN, and ß-lactoglobulin A, respectively), PLSR was best for pH, and LASSO was best for CN micelle size. When traits were divided into 2 classes, SVM had the greatest accuracy for the majority of the traits investigated. Although the well-established PLSR-based method performed competitively, the application of statistical machine learning methods for regression analyses reduced the root mean square error compared with PLSR from between 0.18% (κ-CN) to 3.67% (heat stability). The use of modern statistical machine learning methods for trait prediction from mid-infrared spectroscopy may improve the prediction accuracy for some traits.

Application of machine-learning methods to milk mid-infrared spectra for discrimination of cow milk from pasture or total mixed ration diets.

The prevalence of "grass-fed" labeled food products on the market has increased in recent years, often commanding a premium price. To date, the majority of methods used for the authentication of grass-fed source products are driven by auditing and inspection of farm records. As such, the ability to verify grass-fed source claims to ensure consumer confidence will be important in the future. Mid-infrared (MIR) spectroscopy is widely used in the dairy industry as a rapid method for the routine monitoring of individual herd milk composition and quality. Further harnessing the data from individual spectra offers a promising and readily implementable strategy to authenticate the milk source at both farm and processor levels. Herein, a comprehensive comparison of the robustness, specificity, and accuracy of 11 machine-learning statistical analysis methods were tested for the discrimination of grass-fed versus non-grass-fed milks based on the MIR spectra of 4,320 milk samples collected from cows on pasture or indoor total mixed ration-based feeding systems over a 3-yr period. Linear discriminant analysis and partial least squares discriminant analysis (PLS-DA) were demonstrated to offer the greatest level of accuracy for the prediction of cow diet from MIR spectra. Parsimonious strategies for the selection of the most discriminating wavelengths within the spectra are also highlighted.

Real-time monitoring of quality attributes by in-line Fourier transform infrared spectroscopic sensors at ultrafiltration and diafiltration of bioprocess.

Technologies capable of monitoring product quality attributes and process parameters in real time are becoming popular due to the endorsement of regulatory agencies and also to support the agile development of biotherapeutic pipelines. The utility of vibrational spectroscopic techniques such as Fourier transform mid-infrared (Mid-IR) and multivariate data analysis (MVDA) models allows the prediction of multiple critical attributes simultaneously in real time. This study reports the use of Mid-IR and MVDA model sensors for monitoring of multiple attributes (excipients and protein concentrations) in real time (measurement frequency of every 40 s) at ultrafiltration and diafiltration (UF/DF) unit operation of biologics manufacturing. The platform features integration of fiber optic Mid-IR probe sensors to UF/DF set up at the bulk solution and through a flow cell at the retentate line followed by automated Mid-IR data piping into a process monitoring software platform with pre-loaded partial least square regression (PLS) chemometric models. Data visualization infrastructure is also built-in to the platform so that upon automated PLS prediction of excipients and protein concentrations, the results were projected in a graphical or numerical format in real time. The Mid-IR predicted concentrations of excipients and protein show excellent correlation with the offline measurements by traditional analytical methods. Absolute percent difference values between Mid-IR predicted results and offline reference assay results were ≤5% across all the excipients and the protein of interest; which shows a great promise as a reliable process analytical technology tool.

Authentication and Provenance of Walnut Combining Fourier Transform Mid-Infrared Spectroscopy with Machine Learning Algorithms.

Different varieties and geographical origins of walnut usually lead to different nutritional values, contributing to a big difference in the final price. The conventional analytical techniques have some unavoidable limitations, e.g., chemical analysis is usually time-expensive and labor-intensive. Therefore, this work aims to apply Fourier transform mid-infrared spectroscopy coupled with machine learning algorithms for the rapid and accurate classification of walnut species that originated from ten varieties produced from four provinces. Three types of models were developed by using five machine learning classifiers to (1) differentiate four geographical origins; (2) identify varieties produced from the same origin; and (3) classify all 10 varieties from four origins. Prior to modeling, the wavelet transform algorithm was used to smooth and denoise the spectrum. The results showed that the identification of varieties under the same origin performed the best (i.e., accuracy = 100% for some origins), followed by the classification of four different origins (i.e., accuracy = 96.97%), while the discrimination of all 10 varieties is the least desirable (i.e., accuracy = 87.88%). Our results implicated that using the full spectral range of 700-4350 cm-1 is inferior to using the subsets of the optimal spectral variables for some classifiers. Additionally, it is demonstrated that back propagation neural network (BPNN) delivered the best model performance, while random forests (RF) produced the worst outcome. Hence, this work showed that the authentication and provenance of walnut can be realized effectively based on Fourier transform mid-infrared spectroscopy combined with machine learning algorithms.

Fluorometric atrazine assay based on the use of nitrogen-doped graphene quantum dots and on inhibition of the activity of tyrosinase.

A fluorometric assay is described for the determination of the herbicide atrazine. The assay is based on the use of tyrosinase and fluorescent nitrogen-doped graphene quantum dots (N-GQDs). The N-GQDs were synthesized via one-pot hydrothermal reaction starting from citric acid and ammonia. Their fluorescence excitation and emission maxima are at 355 and 435 nm, and the quantum yield is 18%. Tyrosinase catalyzes the oxidation of dopamine to form dopaquinone which reduces the fluorescence of the N-GQDs through a dynamic quenching process. On addition of atrazine, the catalytic activity of tyrosinase is inhibited. This leads to less formation of dopaquinone and less reduction of fluorescence. The assay has a linear response in the 2.5-100 ng·mL-1 atrazine concentration range, and the detection limit is 1.2 ng·mL-1. The assay was applied to the determination of atrazine in spiked environmental water samples. Graphical abstract Schematic presentation of the fluorometric assay of atrazine detection based on tyrosinase-induced fluorescence (FL) quenching effect on the nitrogen-doped graphene quantum dots (N-GQDs) and inhibitory effect of atrazine on tyrosinase.

Data Fusion of Fourier Transform Mid-Infrared (MIR) and Near-Infrared (NIR) Spectroscopies to Identify Geographical Origin of Wild Paris polyphylla var. yunnanensis.

Origin traceability is important for controlling the effect of Chinese medicinal materials and Chinese patent medicines. Paris polyphylla var. yunnanensis is widely distributed and well-known all over the world. In our study, two spectroscopic techniques (Fourier transform mid-infrared (FT-MIR) and near-infrared (NIR)) were applied for the geographical origin traceability of 196 wild P. yunnanensis samples combined with low-, mid-, and high-level data fusion strategies. Partial least squares discriminant analysis (PLS-DA) and random forest (RF) were used to establish classification models. Feature variables extraction (principal component analysis-PCA) and important variables selection models (recursive feature elimination and Boruta) were applied for geographical origin traceability, while the classification ability of models with the former model is better than with the latter. FT-MIR spectra are considered to contribute more than NIR spectra. Besides, the result of high-level data fusion based on principal components (PCs) feature variables extraction is satisfactory with an accuracy of 100%. Hence, data fusion of FT-MIR and NIR signals can effectively identify the geographical origin of wild P. yunnanensis.

FT-MIR and NIR spectral data fusion: a synergetic strategy for the geographical traceability of Panax notoginseng.

Three data fusion strategies (low-llevel, mid-llevel, and high-llevel) combined with a multivariate classification algorithm (random forest, RF) were applied to authenticate the geographical origins of Panax notoginseng collected from five regions of Yunnan province in China. In low-level fusion, the original data from two spectra (Fourier transform mid-IR spectrum and near-IR spectrum) were directly concatenated into a new matrix, which then was applied for the classification. Mid-level fusion was the strategy that inputted variables extracted from the spectral data into an RF classification model. The extracted variables were processed by iterate variable selection of the RF model and principal component analysis. The use of high-level fusion combined the decision making of each spectroscopic technique and resulted in an ensemble decision. The results showed that the mid-level and high-level data fusion take advantage of the information synergy from two spectroscopic techniques and had better classification performance than that of independent decision making. High-level data fusion is the most effective strategy since the classification results are better than those of the other fusion strategies: accuracy rates ranged between 93% and 96% for the low-level data fusion, between 95% and 98% for the mid-level data fusion, and between 98% and 100% for the high-level data fusion. In conclusion, the high-level data fusion strategy for Fourier transform mid-IR and near-IR spectra can be used as a reliable tool for correct geographical identification of P. notoginseng. Graphical abstract The analytical steps of Fourier transform mid-IR and near-IR spectral data fusion for the geographical traceability of Panax notoginseng.

Genotype, development and tissue-derived variation of cell-wall properties in the lignocellulosic energy crop Miscanthus.

BACKGROUND AND AIMS: Species and hybrids of the genus Miscanthus contain attributes that make them front-runners among current selections of dedicated bioenergy crops. A key trait for plant biomass conversion to biofuels and biomaterials is cell-wall quality; however, knowledge of cell-wall composition and biology in Miscanthus species is limited. This study presents data on cell-wall compositional changes as a function of development and tissue type across selected genotypes, and considers implications for the development of miscanthus as a sustainable and renewable bioenergy feedstock. METHODS: Cell-wall biomass was analysed for 25 genotypes, considering different developmental stages and stem vs. leaf compositional variability, by Fourier transform mid-infrared spectroscopy and lignin determination. In addition, a Clostridium phytofermentans bioassay was used to assess cell-wall digestibility and conversion to ethanol. KEY RESULTS: Important cell-wall compositional differences between miscanthus stem and leaf samples were found to be predominantly associated with structural carbohydrates. Lignin content increased as plants matured and was higher in stem tissues. Although stem lignin concentration correlated inversely with ethanol production, no such correlation was observed for leaves. Leaf tissue contributed significantly to total above-ground biomass at all stages, although the extent of this contribution was genotype-dependent. CONCLUSIONS: It is hypothesized that divergent carbohydrate compositions and modifications in stem and leaf tissues are major determinants for observed differences in cell-wall quality. The findings indicate that improvement of lignocellulosic feedstocks should encompass tissue-dependent variation as it affects amenability to biological conversion. For gene-trait associations relating to cell-wall quality, the data support the separate examination of leaf and stem composition, as tissue-specific traits may be masked by considering only total above-ground biomass samples, and sample variability could be mostly due to varying tissue contributions to total biomass.

Improved soil carbonate determination by FT-IR and X-ray analysis.

In forest soils on calcareous parent material, carbonate is a key component that influences both chemical and physical soil properties and thus fertility and productivity. At low organic carbon contents, it is difficult to distinguish between organic and inorganic carbon, e.g. carbonates, in soils. The common gas-volumetric method to determine carbonate has a number of disadvantages. We hypothesize that a combination of two spectroscopic methods, which account for different forms of carbonate, can be used to model soil carbonate in our region. Fourier transform mid-infrared spectroscopy was combined with X-ray diffraction to develop a model based on partial least squares regression. Results of the gas-volumetric Scheibler method were corrected for the calcite/dolomite ratio. The best model performance was achieved when we combined the two analytical methods using four principal components. The root mean squared error of prediction decreased from 13.07 to 11.57, while full cross-validation explained 94.5 % of the variance of the carbonate content. This is the first time that a combination of the proposed methods has been used to predict carbonate in forest soils, offering a simple method to precisely estimate soil carbonate contents while increasing accuracy in comparison with spectroscopic approaches proposed earlier. This approach has the potential to complement or substitute gas-volumetric methods, specifically in study areas with low soil heterogeneity and similar parent material or in long-term monitoring by consecutive sampling.

FT-MIR supported Electrical Impedance Spectroscopy based study of sugar adulterated honeys from different floral origin.

This study sought to detect the presence of sucrose as an adulterant in selected honey varieties from different floral origins by employing Electrical Impedance Spectroscopy (EIS) technique which has been simultaneously supported by Fourier Transform-Mid Infrared Spectroscopy (FT-MIR) measurements to provide a rapid, robust yet simple platform for honey quality evaluation. Variation of electrical parameters such as impedance, capacitance and conductance for 10%, 20%, 30%, 40%, 50%, 60% and 70% (w/w) sucrose syrup (SS) adulterated honey samples are analyzed and their respective current-voltage (I-V) characteristics are studied. Capacitance, conductance and net current flowing through the system are observed to decrease linearly whereas system impedance has been found to increase similarly with the increase in adulterant content. Also, FT-MIR measurements in the spectral region between 1800cm-1 and 650cm-1 reveal the increment of absorbance values due to the addition of SS. Full-Width-at-Half-Maximum (FWHM) is estimated from the spectral peak 1056cm-1 for all pure and adulterated honey samples and is observed to be linearly increasing with increase in adulterant content. Finally, the coefficient of sensitivity has been extracted for all varieties of honey considered in terms of the measured conductance values.

Evaluation of a newly developed mid-infrared sensor for real-time monitoring of yeast fermentations.

A mid-infrared (MIR) sensor using the attenuated total reflection (ATR) technique has been developed for real-time monitoring in biotechnology. The MIR-ATR sensor consists of an IR emitter as light source, a zinc selenide ATR prism as boundary to the process, and four thermopile detectors, each equipped with an optical bandpass filter. The suitability of the sensor for practical application was tested during aerobic batch-fermentations of Saccharomyces cerevisiae by simultaneous monitoring of glucose and ethanol. The performance of the sensor was compared to a commercial Fourier transform mid-infrared (FT-MIR) spectrometer by on-line measurements in a bypass loop. Sensor and spectrometer were calibrated by multiple linear regression (MLR) in order to link the measured absorbance in the transmission ranges of the four optical sensor channels to the analyte concentrations. For reference analysis, high-performance liquid chromatography (HPLC) was applied. Process monitoring using the sensor yielded in standard errors of prediction (SEP) of 6.15 g/L and 1.36 g/L for glucose and ethanol. In the case of the FT-MIR spectrometer the corresponding SEP values were 4.34 g/L and 0.61 g/L, respectively. The advantages of optical multi-channel mid-infrared sensors in comparison to FT-MIR spectrometer setups are the compactness, easy process implementation and lower price.

Improving the prediction ability of FT-MIR spectroscopy to assess titratable acidity in cow's milk.

This study investigated the potential application of Fourier transform mid-infrared spectroscopy (FT-MIR) for the determination of titratable acidity (TA) in cow's milk. The prediction model was developed on 201 samples collected from cows in early and late lactation, and was successively used to predict TA on samples collected from cows in early lactation and in samples with high somatic cell count. The root mean square error of cross-validation of the model by using external validation dataset was 0.09 °Soxhlet-Henkel/50 mL. Applying the model on milk samples from cows in early lactation or with high somatic cell count, the root mean square error of prediction was 0.163 °Soxhlet-Henkel/50 mL, with a RER and RPD of 23.9 and 5.1, respectively. Our results seem to indicate that FT-MIR can be used in individual milk samples to accurately predict TA, and has the potential to be adopted to measure routinely the TA of milk.

Combination of visible and mid-infrared spectra for the prediction of chemical parameters of wines.

Rapid and environmentally friendly methods for the prediction of chemical compositions have been an interest in the wine industry. The objective of the study was to show the potentials of combined use of visible and mid-infrared (MIR) spectroscopies to improve the prediction of various chemical compounds of wine as opposed to using mid-infrared range only. Wine samples of twelve grape varieties from two harvest years were analyzed. The chemical composition of wine samples was related to MIR and visible spectra using orthogonal partial least square (OPLS) regression technique. The prediction abilities were tested with crossvalidation and independent validation sets. The coefficient of determination of validation (R2val) for anthocyanin compounds of red wines were between 0.76 and 0.90, and that for total phenol content was 0.90. Range of R2val for glycerol, glycerol/ethanol ratio, malic acid, o-coumaric acid and °Brix were between 0.77 and 0.96. The spectral ranges that played significant roles in the predictions were also determined. The validations with independent data sets showed that the combination of visible and MIR ranges with multivariate methods improved the prediction of anthocyanin compounds and total phenols; produced comparable results for the rest of the parameters as MIR. This is the first study in the literature that shows the practical use of visible spectra along MIR. The combined use of these spectral ranges with multivariate models can be applied for the rapid, on-line determination of quality parameters and chemical profiles of wines.

Rapid determination of phospholipid content of vegetable oils by FTIR spectroscopy combined with partial least-square regression.

A rapid mid-FTIR method was developed to quantitatively determine the total phospholipid (PL) content of vegetable oils. The method simply requires that the oil be diluted 4:1 (w/w) with hexane, its spectrum taken and ratioed against a hexane background. A calibration was devised using partial least squares by adding purified soybean PL at levels of 0.02-2.0% to phospholipid-free oils (soybean, rapeseed, sunflower) using the spectral region encompassing 1,357-1,000 cm(-1) and validated using the AOCS 12-55. Using calibration and leave-one-out cross-validation predictive errors, a 200-20,000 ppm calibration was accurate to within ± 362 and 488 ppm, respectively, while for sub-calibrations ranging from 200 to 2000; 2000 to 8000 and 8000 to 20,000 ppm, they were ± 72-172, ± 119-220, and ± 242-371 ppm, respectively. Although limited to 3 oil types in this study, the calibration is simple to devise and can be broadened to the universe of oil types of interest, the analytical protocol being straightforward and the analysis readily automatable.