Reproducibility and repeatability of measures of milk coagulation properties and predictive ability of mid-infrared reflectance spectroscopy

The objectives of the study were to estimate the reproducibility and repeatability of milk coagulation properties (MCP) measured by a computerized renneting meter (CRM) and to evaluate the predictive ability of mid-infrared spectroscopy (MIRS) as an innovative technology for the assessment of rennet coagulation time (RCT, min) and curd firmness (a₃₀, mm). Four samples without addition of preservative (NP) and 4 samples with Bronopol addition (PS) were collected from each of 83 Holstein-Friesian cows. Six hours after collection, 2 replicated measures of MCP were obtained with CRM using 1 NP and 1 PS sample from each cow. Mid-infrared spectra of the remaining NP and PS samples from each animal were recorded after 6 h, 4 d, and 8 d after sampling. Two groups of calibration equations were developed using MIRS spectra and CRM measures of MCP as reference data obtained from analysis of NP and PS, respectively. Reproducibility and repeatability of CRM measures were obtained from REML estimation of variance components on the basis of a linear model including the fixed effects of herd and days in milk class and the random effects of cows, sample treatment (addition or no addition of preservative), and the interaction between cow and sample treatment. Coefficient of reproducibility is an indicator of the agreement between 2 measurements of MCP for the same milk sample preserved with or without addition of Bronopol. Coefficient of repeatability is an indicator of the agreement between repeated measures of MCP. Pearson correlations between MCP measures for NP and PS were 0.97 and 0.83 for RCT and a₃₀, respectively. Reproducibility of CRM measures under different preserving conditions of milk was 93.5% for RCT and 64.6% for a₃₀. Repeatabilities of RCT and a₃₀ measures were 95.7 and 77.3%, respectively. Based on the estimated cross-validation standard errors and coefficients of determination and ratios of standard errors of cross-validation to standard deviation of reference data, the predictive ability of MIRS calibration equations was moderate for RCT and unsatisfactory for a_30. Predictive ability of equations based on spectra and MCP measures of PS was greater than that of equations based on data of NP. The study did not provide conclusive evidence on the effectiveness of MIRS as a predictive tool for MCP and it requires an enlargement of the variability of milk sampling circumstances. Because the relevance of MIRS predictions in relation to breeding programs for MCP based on indicator traits relies on the genetic variation of MIRS predictions and on phenotypic and genetic correlations between MIRS predictions and MCP measures, additional specific investigations on these topics are needed.     

Predicting dry matter intake in Canadian Holstein dairy cattle using milk mid-infrared reflectance spectroscopy and other commonly available predictors via artificial neural networks

Dry matter intake (DMI) is a fundamental component of the animal's feed efficiency, but measuring DMI of individual cows is expensive. Mid-infrared reflectance spectroscopy (MIRS) on milk samples could be an inexpensive alternative to predict DMI. The objectives of this study were (1) to assess if milk MIRS data could improve DMI predictions of Canadian Holstein cows using artificial neural networks (ANN); (2) to investigate the ability of different ANN architectures to predict unobserved DMI; and (3) to validate the robustness of developed prediction models. A total of 7,398 milk samples from 509 dairy cows distributed over Canada, Denmark, and the United States were analyzed. Data from Denmark and the United States were used to increase the training data size and variability to improve the generalization of the prediction models over the lactation. For each milk spectra record, the corresponding weekly average DMI (kg/d), test-day milk yield (MY, kg/d), fat yield (FY, g/d), and protein yield (PY, g/d), metabolic body weight (MBW), age at calving, year of calving, season of calving, days in milk, lactation number, country, and herd were available. The weekly average DMI was predicted with various ANN architectures using 7 predictor sets, which were created by different combinations MY, FY, PY, MBW, and MIRS data. All predictor sets also included age of calving and days in milk. In addition, the classification effects of season of calving, country, and lactation number were included in all models. The explored ANN architectures consisted of 3 training algorithms (Bayesian regularization, Levenberg-Marquardt, and scaled conjugate gradient), 2 types of activation functions (hyperbolic tangent and linear), and from 1 to 10 neurons in hidden layers). In addition, partial least squares regression was also applied to predict the DMI. Models were compared using cross-validation based on leaving out 10% of records (validation A) and leaving out 10% of cows (validation B). Superior fitting statistics of models comprising MIRS information compared with the models fitting milk, fat and protein yields suggest that other unknown milk components may help explain variation in weekly average DMI. For instance, using MY, FY, PY, and MBW as predictor variables produced a predictive accuracy (r) ranging from 0.510 to 0.652 across ANN models and validation sets. Using MIRS together with MY, FY, PY, and MBW as predictors resulted in improved fitting (r = 0.679–0.777). Including MIRS data improved the weekly average DMI prediction of Canadian Holstein cows, but it seems that MIRS predicts DMI mostly through its association with milk production traits and its utility to estimate a measure of feed efficiency that accounts for the level of production, such as residual feed intake, might be limited and needs further investigation. The better predictive ability of nonlinear ANN compared with linear ANN and partial least squares regression indicated possible nonlinear relationships between weekly average DMI and the predictor variables. In general, ANN using Bayesian regularization and scaled conjugate gradient training algorithms yielded slightly better weekly average DMI predictions compared with ANN using the Levenberg-Marquardt training algorithm.     

Predicting the dry matter intake of grazing dairy cows using infrared reflectance spectroscopy analysis

The objective of this study was to compare mid-infrared reflectance spectroscopy (MIRS) analysis of milk and near-infrared reflectance spectroscopy (NIRS) analysis of feces with regard to their ability to predict the dry matter intake (DMI) of lactating grazing dairy cows. A data set comprising 1,074 records of DMI from 457 cows was available for analysis. Linear regression and partial least squares regression were used to develop the equations using the following variables: (1) milk yield (MY), fat percentage, protein percentage, body weight (BW), stage of lactation (SOL), and parity (benchmark equation); (2) MIRS wavelengths; (3) MIRS wavelengths, MY, fat percentage, protein percentage, BW, SOL, and parity; (4) NIRS wavelengths; (5) NIRS wavelengths, MY, fat percentage, protein percentage, BW, SOL, and parity; (6) MIRS and NIRS wavelengths; and (7) MIRS wavelengths, NIRS wavelengths, MY, fat percentage, protein percentage, BW, SOL, and parity. The equations were validated both within herd using animals from similar experiments and across herds using animals from independent experiments. The accuracy of equations was greater for within-herd validation compared with across-herds validation. Across-herds validation was deemed the more suitable method to assess equations for robustness and real-world application. The benchmark equation was more accurate [coefficient of determination (R²) = 0.60; root mean squared error (RMSE) = 1.68 kg] than MIRS alone (R² = 0.30; RMSE = 2.23 kg) or NIRS alone (R² = 0.16; RMSE = 2.43 kg). The combination of the benchmark equation with MIRS (R² = 0.64; RMSE = 1.59 kg) resulted in slightly superior fitting statistics compared with the benchmark equation alone. The combination of the benchmark equation with NIRS (R² = 0.58; RMSE = 1.71 kg) did not result in a more accurate prediction equation than the benchmark equation. The combination of MIRS and NIRS wavelengths resulted in superior fitting statistics compared with either method alone (R² = 0.36; RMSE = 2.15 kg). The combination of the benchmark equation and MIRS and NIRS wavelengths resulted in the most accurate equation (R² = 0.68; RMSE = 1.52 kg). A further analysis demonstrated that Holstein-Friesian cows could predict the DMI of Jersey × Holstein-Friesian crossbred cows using both MIRS and NIRS. Similarly, the Jersey × Holstein-Friesian animals could predict the DMI of Holstein-Friesian cows using both MIRS and NIRS. The equations developed in this study have the capacity to predict DMI of grazing dairy cows. From a practicality perspective, MIRS in combination with variables in the benchmark equation is the most suitable equation because MIRS is currently used on all milk-recorded milk samples from dairy cows.     

Prediction of fatty acid profiles in cow,ewe, and goat milk by mid-infrared spectrometry

Mid-infrared (MIR) spectrometry was used to estimate the fatty acid (FA) composition in cow, ewe, and goat milk. The objectives were to compare different statistical approaches with wavelength selection to predict the milk FA composition from MIR spectra, and to develop equations for FA in cow, goat, and ewe milk. In total, a set of 349 cow milk samples, 200 ewe milk samples, and 332 goat milk samples were both analyzed by MIR and by gas chromatography, the reference method. A broad FA variability was ensured by using milk from different breeds and feeding systems. The methods studied were partial least squares regression (PLS), first-derivative pretreatment + PLS, genetic algorithm + PLS, wavelets + PLS, least absolute shrinkage and selection operator method (LASSO), and elastic net. The best results were obtained with PLS, genetic algorithm + PLS and first derivative + PLS. The residual standard deviation and the coefficient of determination in external validation were used to characterize the equations and to retain the best for each FA in each species. In all cases, the predictions were of better quality for FA found at medium to high concentrations (i.e., for saturated FA and some monounsaturated FA with a coefficient of determination in external validation >0.90). The conversion of the FA expressed in grams per 100 mL of milk to grams per 100 g of FA was possible with a small loss of accuracy for some FA.     

Could predicting fatty acid profile by mid-infrared reflectance spectroscopy be used as a method to increase the value added by milk production chains?

The aims of this work were (1) to develop prediction equations from mid-infrared spectroscopy (MIRS) to establish a detailed fatty acid (FA) composition of milk; (2) to propose a milk FA index, utilizing MIRS-developed equations, in which the precision of the FA-prediction equations is taken into account to increase the value of milk; and (3) to show application examples. A total of 651 bulk cow milk samples were collected from 245 commercial farms in northwest Italy. The results of the 651 gas chromatography analyses were used to establish (421 samples) and to validate (230 samples) the outcomes of the FA composition prediction that had been obtained by MIRS. A class-based approach, in which the obtained MIRS equations were used, was proposed to define a milk classification. The method provides a numerical index [milk FA index (MFAI)] that allows a premium price to be quantified to increase the value of a favorable FA profile of milk. Ten FA were selected to calculate MFAI, according to their relevance for human health and potential cheese sensory properties, and animal welfare and environmental sustainability were also considered. These factors were selected as dimensions of MFAI. A statistical analysis and expert judgment aggregation were performed on the selected FA by weighting the FA and normalizing the dimensions to reduce redundancy. A class approach was applied, using the precision of the MIRS equations to establish the classes. The median FA concentration of the data set was set as a reference value of class 0. The width, number, and limits of classes above and below the median were calculated using the 95% confidence level of the standard error of prediction, corrected with the bias of each FA. A progressive number and a positive or negative sign were assigned to each FA class above or below the median according to their role in the above mentioned dimensions. The sum of the numbers of each class, associated with its sign for each FA, was used to generate MFAI. The MFAI was applied to dairy farms characterized by different feeding strategies, all of which deliver milk to a commercial dairy plant. The MFAI values ranged from 0.7 to 4.2, and large variations, which depended on the cows' diet and forage quality, were observed for each feeding system. The proposed method has been found to be flexible and adaptable to several contexts on both intensive and extensive dairy farms.     

Effectiveness of mid-infrared spectroscopy for the prediction of cow milk metabolites

Proton nuclear magnetic resonance (¹H NMR) spectroscopy is acknowledged as one of the most powerful analytical methods with cross-cutting applications in dairy foods. To date, the use of ¹H NMR spectroscopy for the collection of milk metabolic profile is hindered by costly and time-consuming sample preparation and analysis. The present study aimed at evaluating the accuracy of mid-infrared spectroscopy (MIRS) as a rapid method for the prediction of cow milk metabolites determined through ¹H NMR spectroscopy. Bulk milk (n = 72) and individual milk samples (n = 482) were analyzed through one-dimensional ¹H NMR spectroscopy and MIRS. Nuclear magnetic resonance spectroscopy identified 35 milk metabolites, which were quantified in terms of relative abundance, and MIRS prediction models were developed on the same 35 milk metabolites, using partial least squares regression analysis. The best MIRS prediction models were developed for galactose-1-phosphate, glycerophosphocholine, orotate, choline, galactose, lecithin, glutamate, and lactose, with coefficient of determination in external validation from 0.58 to 0.85, and ratio of performance to deviation in external validation from 1.50 to 2.64. The remaining 27 metabolites were poorly predicted. This study represents a first attempt to predict milk metabolome. Further research is needed to specifically address whether developed prediction models may find practical application in the dairy sector, with particular regard to the screening of dairy cows' metabolic status, the quality control of dairy foods, and the identification of processed milk or incorrectly stored milk.     

Total antioxidant activity of bovine milk: Phenotypic variation and predictive ability of mid-infrared spectroscopy

The phenotypic variation of milk total antioxidant activity (TAA) and the ability of mid-infrared spectroscopy to predict this novel trait was investigated. Total antioxidant activity was measured through the reference spectrophotometric method on 1249 individual milk samples of Holstein Friesian cows. Sources of variation of milk TAA were investigated using a mixed model, which included the fixed effects of days in milk, parity and calving season, and the random effects of herd-test-date and error. Mid-infrared spectroscopy prediction models were developed using partial least squares regression approach. The average level of milk TAA was 6.93 mmoL L⁻¹ of Trolox equivalents; this exhibited a coefficient of variation of 15% and showed weak phenotypic correlations with milk quality traits. Values of TAA were lower in early lactation than in late lactation. Mid-infrared spectroscopy prediction models reached a coefficient of determination in external validation of 0.41, suggesting that they are not adequate for analytical purposes.     

Genetic components of milk Fourier-transform infrared spectra used to predict breeding values for milk composition and quality traits in dairy goats

The usual practice today is that milk component phenotypes are predicted using Fourier-transform infrared (FTIR) spectra and they are then, together with pedigree information, used in BLUP for calculation of individual estimated breeding values. Here, this is referred to as the indirect prediction (IP) approach. An alternative approach—a direct prediction (DP) method—is proposed, where genetic analyses are directly conducted on the milk FTIR spectral variables. Breeding values of all derived milk traits (protein, fat, fatty acid composition, and coagulation properties, among others) can then be predicted as traits correlated only to the genetic information of the spectra. For the DP, no need exists to predict the phenotypes before calculating breeding values for each of the traits—the genetic analysis is done once for the spectra, and is applicable to all traits derived from the spectra. The aim of the study was to compare the effects of DP and IP of milk composition and quality traits on prediction error variance (PEV) and genetic gain. A data set containing 27,927 milk FTIR spectral observations and milk composition phenotypes (fat, lactose, and protein) belonging to 14,869 goats of 271 herds was used for training and evaluating models. Partial least squares regression was used for calibrating prediction models for fat, protein, and lactose percentages. Restricted maximum likelihood was used to estimate variance components of the spectral variables after principal components analysis was applied to reduce the spectral dimension. Estimated breeding values were predicted for fat, lactose, and protein percentages using DP and IP methods. The DP approach reduced the mean PEV by 3.73, 4.07, and 7.04% for fat, lactose, and protein percentages, respectively, compared with the IP method. Given the reduction in PEV, relative genetic gains were 2.99, 2.78, and 4.85% for fat, lactose, and protein percentages, respectively. We concluded that more accurate estimated breeding values could be found using genetic components of milk FTIR spectra compared with single-trait animal model analyses on phenotypes predicted from the spectra separately. The potential and application is not only limited to milk FTIR spectra, but could also be extended to any spectroscopy techniques implemented in other species and for other traits.     

Effectiveness of mid-infrared spectroscopy for the prediction of detailed protein composition and contents of protein genetic variants of individual milk of Simmental cows

Mid-infrared (MIR) spectroscopy was used to predict the detailed protein composition of 1,517 milk samples of Simmental cows. Contents of milk protein fractions and genetic variants were quantified by reversed-phase HPLC. The most accurate predictions were those obtained for total protein, casein (CN), α_S1-CN, β-lactoglobulin (LG), glycosylated κ-CN, and whey protein content, which exhibited coefficients of determination between predicted and measured values in cross-validation (1-VR) ranging from 0.61 to 0.78. Less favorable were results for β-CN (1-VR = 0.53), α_S2-CN, and κ-CN (1-VR = 0.49). Neither the content of α-LA nor that of γ-CN was accurately predicted by MIR. Predicting the content of the most common milk protein genetic variants (κ-CN A and B; β-CN A¹, A², and B; and β-LG A and B) was unfeasible (1-VR <0.15 for the content of κ-CN genetic variants and 1-VR <0.01 for the content of β-CN variants). The best predictions were obtained for β-LG A and β-LG B contents (1-VR of 0.60 and 0.44, respectively). Results indicated that MIR is not applicable for predicting individual milk protein composition with high accuracy. However, MIR spectroscopy predictions may play a role as indicator traits in selective breeding to enhance milk protein composition. The genetic correlation between MIR spectroscopy predictions and measures of milk protein composition needs to be investigated, as it affects the suitability of MIR spectroscopy predictions as indicator traits in selective breeding.     

Prediction of blood metabolites from milk mid-infrared spectra in early-lactation cows

Dairy cows commonly experience an unbalanced energy status in early lactation, and this condition can lead to the onset of several metabolic disorders. Blood metabolic profile testing is a valid tool to monitor and detect the most common early lactation disorders, but blood sampling and analysis are time-consuming and expensive, and the procedure is invasive and stressful for the cows. Mid-infrared (MIR) spectroscopy is routinely used to analyze milk composition, being a cost-effective and nondestructive method. The present study aimed to assess the feasibility of using routine milk MIR spectra for the prediction of main blood metabolites in dairy cows, and to investigate associations between measured blood metabolites and milk traits. Twenty herds of Holstein Friesian, Brown Swiss, or Simmental cows located in Northeast Italy were visited 1 to 4 times between December 2017 and June 2018, and blood and milk samples were collected from all lactating cows within 35 d in milk. Concentrations of main blood metabolites and milk MIR spectra were recorded from 295 blood and milk samples and used to develop prediction models for blood metabolic traits through backward interval partial least squares analysis. Blood β-hydroxybutyrate (BHB), urea, and nonesterified fatty acids were the most predictable traits, with coefficients of determination of 0.63, 0.58, and 0.52, respectively. On the contrary, predictive performance for blood glucose, triglycerides, cholesterol, glutamic oxaloacetic transaminase, and glutamic pyruvic transaminase were not accurate. Associations of blood BHB and urea with their respective contents in milk were moderate to strong, whereas all other correlations were weak. Predicted blood BHB showed an improved performance in detecting cows with hyperketonemia (blood BHB ≥ 1.2 mmol/L), compared with commercial calibration equation for milk BHB. Results highlighted the opportunity of using milk MIR spectra to predict blood metabolites and thus to collect routine information on the metabolic status of early-lactation cows at a population level.     

Invited review: Corrected milk: Reconsideration of common equations and milk energy estimates

Corrected milk equations were developed in attempts to bring milk weights to a standardized basis for comparison by expressing the weight and composition of milk as corrected to the energy content of milk of a specific composition. Expressed as milk weights familiar on farm and in commerce, this approach integrates energy contributions of the dissimilar components to make the mass units more comparable. Such values are applied in evaluating feed efficiency, lactation performance, and global milk production, as functional units for lifecycle assessments, and in translation of research results. Corrected milk equations are derived from equations relating milk gross energy to milk composition. First, a milk energy equation is used to calculate the energy value of the milk composition to correct to (e.g., 0.695 Mcal/kg for milk with 3.5% fat, 3.05% true protein, and 4.85% lactose). That energy value is divided into the energy equation to give the corrected milk equation. Confusion has arisen, as different equations purport to correct to the same milk composition; their differences are based on uses of different energy equations or divisors. Accuracy of corrected milk equations depends on the accuracy of the energy equations used to create them. Energy equations have evolved over time as different milk component analyses have become more available. Inclusion of multiple milk components more accurately predicts milk energy content than does fat content alone. Omission of components from an equation requires the assumption that their content in milk is constant or highly correlated with an included component. Neither of these assumptions is true. Milk energy equations evaluated on a small data set of measured milk values have demonstrated that equations that incorporate protein, fat, and lactose contents multiplied by the gross energy of each component more closely predict milk energy than equations containing fewer components or regression-derived equations. This provides a tentative recommendation for using energy equations that include the 3 main milk components and their gross energy multipliers for predicting milk energy and deriving corrected milk equations. Accuracy of energy equations is affected by the accuracy of gross energy values of individual components and variability of milk composition. Lactose has consistent reported gross energy values. In contrast, gross energy of milk fat and protein vary as their compositional profiles change. Future refinements could assess accuracy of milk fat and protein gross energy and whether that appreciably improves milk energy predictions. Fat gross energy has potential to be calculated using the milk fatty acid profile, although the influence on gross energy may be small. For research, direct reporting of milk energy values, rather than corrected milk, provides the most explicit, least manipulated form of the data. However, provision of corrected milk values in addition to information on components can serve to translate the energy information to a form familiar to and widely used in the field. When reporting corrected milk data, the corrected milk equation, citation for the energy equation used, and composition and energy contents of the corrected milk must be described to make clear what the values represent.     

Predicting pregnancy status from mid-infrared spectroscopy in dairy cow milk using deep learning

Accurately identifying pregnancy status is imperative for a profitable dairy enterprise. Mid-infrared (MIR) spectroscopy is routinely used to determine fat and protein concentrations in milk samples. Mid-infrared spectra have successfully been used to predict other economically important traits, including fatty acid content, mineral content, body energy status, lactoferrin, feed intake, and methane emissions. Machine learning has been used in a variety of fields to find patterns in vast quantities of data. This study aims to use deep learning, a sub-branch of machine learning, to establish pregnancy status from routinely collected milk MIR spectral data. Milk spectral data were obtained from National Milk Records (Chippenham, UK), who collect large volumes of data continuously on a monthly basis. Two approaches were followed: using genetic algorithms for feature selection and network design (model 1), and transfer learning with a pretrained DenseNet model (model 2). Feature selection in model 1 showed that the number of wave points in MIR data could be reduced from 1,060 to 196 wave points. The trained model converged after 162 epochs with validation accuracy and loss of 0.89 and 0.18, respectively. Although the accuracy was sufficiently high, the loss (in terms of predicting only 2 labels) was considered too high and suggested that the model would not be robust enough to apply to industry. Model 2 was trained in 2 stages of 100 epochs each with spectral data converted to gray-scale images and resulted in accuracy and loss of 0.97 and 0.08, respectively. Inspection on inference data showed prediction sensitivity of 0.89, specificity of 0.86, and prediction accuracy of 0.88. Results indicate that milk MIR data contains features relating to pregnancy status and the underlying metabolic changes in dairy cows, and such features can be identified by means of deep learning. Prediction equations from trained models can be used to alert farmers of nonviable pregnancies as well as to verify conception dates.     

Real-time milk analysis integrated with stacking ensemble learning as a tool for the daily prediction of cheese-making traits in Holstein cattle

Cheese-making traits in dairy cattle are important to the dairy industry but are difficult to measure at the individual level because there are limitations on collecting phenotypic information. Mid-infrared spectroscopy has its advantages, but it can only be used during monthly milk recordings. Recently, in-line devices for real-time analysis of milk quality have been developed. The AfiLab recording system (Afimilk) offers significant benefits as phenotypes can be collected from each cow at each milking session. The objective of this study was to assess the potential of integrating AfiLab real-time milk analyzer measures with the stacking ensemble learning technique using heterogeneous base learners for the in-line daily monitoring of cheese-making traits in Holstein cattle with a view to developing a precision livestock farming system for monitoring the technological quality of milk. Data and samples for wet-laboratory analyses were collected from 499 Holstein cows belonging to 2 farms where the AfiLab system was installed. The traits of concern were 9 milk coagulation traits [3 milk coagulation properties (MCP), and 6 curd firming traits (CF_t)], and 7 cheese-making traits [3 cheese yield (CY) traits, and 4 milk nutrient recovery in the curd (REC) traits]. The near-infrared AfiLab spectral data and on-farm information (days in milk and parity) were used to assess the predictive ability of different statistical methods [elastic net (EN), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), and artificial neural network (ANN)] across different cross-validation scenarios. These statistical methods were considered the base learners, which were then combined in a stacking ensemble learning. Results indicate that including information on the cows (days in milk and parity) in the AfiLab infrared prediction increased its accuracy by 10.3% for traditional MCP, 13.8% for curd firming, 9.8% for CY, and 11.2% for REC traits compared with those obtained from near-infrared AfiLab alone. The statistical approaches exhibited high prediction accuracies (R²) averaged across the cross-validation scenarios for traditional MCP (0.58 for ANN, 0.55 for EN and GBM, 0.52 for XGBoost, and 0.62 for stacking ensemble), CF_t (0.55 for ANN, 0.54 for EN and GBM, 0.53 for XGBoost, and 0.61 for stacking ensemble), and similar R² averages for CY and REC (0.55 for ANN, 0.54 for EN and GBM, 0.53 for XGBoost, and 0.61 for stacking ensemble). The ANN approach was more accurate than the other base learners (EN, GBM, and XGBoost) and improved accuracy across cross-validation scenarios on average by 7% for traditional MCP, 5% for CF_t, 8% for CY, and 7% for REC. The stacking ensemble method improved prediction accuracy by 3% to 31% for traditional MCP, 2% to 26% for CF_t, 1% to 38% for CY traits, and 2% to 27% for REC traits compared with the base learners. The prediction accuracies of the different approaches evaluated tended to decrease from the 10-fold cross-validation to the independent validation scenario, although there was a smaller reduction in prediction accuracy with the stacking ensemble learning technique across all the cross-validation scenarios. Our results show that combining in-line on-farm information with stacking ensemble machine learning represents an effective alternative for obtaining robust daily predictions of milk cheese-making traits.     

Prediction and repeatability of milk coagulation properties and curd-firming modeling parameters of ovine milk using Fourier-transform infrared spectroscopy and Bayesian models

The aim of this study was to apply Bayesian models to the Fourier-transform infrared spectroscopy spectra of individual sheep milk samples to derive calibration equations to predict traditional and modeled milk coagulation properties (MCP), and to assess the repeatability of MCP measures and their predictions. Data consisted of 1,002 individual milk samples collected from Sarda ewes reared in 22 farms in the region of Sardinia (Italy) for which MCP and modeled curd-firming parameters were available. Two milk samples were taken from 87 ewes and analyzed with the aim of estimating repeatability, whereas a single sample was taken from the other 915 ewes. Therefore, a total of 1,089 analyses were performed. For each sample, 2 spectra in the infrared region 5,011 to 925 cm^?1 were available and averaged before data analysis. BayesB models were used to calibrate equations for each of the traits. Prediction accuracy was estimated for each trait and model using 20 replicates of a training-testing validation procedure. The repeatability of MCP measures and their predictions were also compared. The correlations between measured and predicted traits, in the external validation, were always higher than 0.5 (0.88 for rennet coagulation time). We confirmed that the most important element for finding the prediction accuracy is the repeatability of the gold standard analyses used for building calibration equations. Repeatability measures of the predicted traits were generally high (≥95%), even for those traits with moderate analytical repeatability. Our results show that Bayesian models applied to Fourier-transform infrared spectra are powerful tools for cheap and rapid prediction of important traits in ovine milk and, compared with other methods, could help in the interpretation of results.     

Precalibration evaluation procedures for mid-infrared milk analyzers

The purpose of this paper is to present a detailed account of the precalibration procedures developed and implemented by the USDA Federal Milk Market Administrators (FMMA) for evaluating mid-infrared (MIR) milk analyzers. Mid-infrared analyzers specifically designed for milk testing provide a rapid and cost-effective means for determining milk composition for payment and dairy herd improvement programs. These instruments determine the fat, protein, and lactose content of milk, and enable the calculation of total solids, solids-not-fat, and other solids. All MIR analyzers are secondary testing instruments that require calibration by chemical reference methods. Precalibration is the process of assuring that the instrument is in good working order (mechanically and electrically) and that the readings before calibration are stable and optimized. The main components of precalibration are evaluation of flow system integrity, homogenization efficiency, water repeatability, zero shift, linearity, primary slope, milk repeatability, purging efficiency, and establishment of intercorrection factors. These are described in detail and apply to both filter-based and Fourier transform infrared instruments operating using classical primary and reference wavelengths. Under the USDA FMMA Precalibration Evaluation Program, the precalibration procedures were applied longitudinally over time using a wide variety of instruments and instrument models. Instruments in this program were maintained to pass the criteria for all precalibration procedures. All instruments used similar primary wavelengths to measure fat, protein, and lactose but there were differences in reference wavelength selection. Intercorrection factors were consistent over time within all instruments and similar among groups of instruments using similar primary and reference wavelengths. However, the magnitude and sign of the intercorrection factors were significantly affected by the choice of reference wavelengths.     

Short communication: Prediction of milk coagulation and acidity traits in Mediterranean buffalo milk using Fourier-transform mid-infrared spectroscopy

Milk coagulation and acidity traits are important factors to inform the cheesemaking process. Those traits have been deeply studied in bovine milk, whereas scarce information is available for buffalo milk. However, the dairy industry is interested in a method to determine milk coagulation and acidity features quickly and in a cost-effective manner, which could be provided by Fourier-transform mid-infrared (FT-MIR) spectroscopy. The aim of this study was to evaluate the potential of FT-MIR to predict coagulation and acidity traits of Mediterranean buffalo milk. A total of 654 records from 36 herds located in central Italy with information on milk yield, somatic cell score, milk chemical composition, milk acidity [pH, titratable acidity (TA)], and milk coagulation properties (rennet coagulation time, curd firming time, and curd firmness) were available for statistical analysis. Reference measures of milk acidity and coagulation properties were matched with milk spectral information, and FT-MIR prediction models were built using partial least squares regression. The data set was divided into a calibration set (75%) and a validation set (25%). The capacity of FT-MIR spectroscopy to correctly classify milk samples based on their renneting ability was evaluated by a canonical discriminant analysis. Average values for milk coagulation traits were 13.32 min, 3.24 min, and 39.27 mm for rennet coagulation time, curd firming time, and curd firmness, respectively. Milk acidity traits averaged 6.66 (pH) and 7.22 Soxhlet-Henkel degrees/100 mL (TA). All milk coagulation and acidity traits, except for pH, had high variability (17 to 46%). Prediction models of coagulation traits were moderately to scarcely accurate, whereas the coefficients of determination of external validation were 0.76 and 0.66 for pH and TA, respectively. Canonical discriminant analysis indicated that information on milk coagulating ability is present in the MIR spectra, and the model correctly classified as noncoagulating the 91.57 and 67.86% of milk samples in the calibration and validation sets, respectively. In conclusion, our results can be relevant to the dairy industry to classify buffalo milk samples before processing.     

Mid-infrared prediction of bovine milk fatty acids across multiple breeds, production systems, and countries

Increasing consumer concern exists over the relationship between food composition and human health. Because of the known effects of fatty acids on human health, the development of a quick, inexpensive, and accurate method to directly quantify the fatty acid (FA) composition in milk would be valuable for milk processors to develop a payment system for milk pertinent to their customer requirements and for farmers to adapt their feeding systems and breeding strategies accordingly. The aim of this study was (1) to confirm the ability of mid-infrared spectrometry (MIR) to quantify individual FA content in milk by using an innovative procedure of sampling (i.e., samples were collected from cows belonging to different breeds, different countries, and in different production systems); (2) to compare 6 mathematical methods to develop robust calibration equations for predicting the contents of individual FA in milk; and (3) to test interest in using the FA equations developed in milk as basis to predict FA content in fat without corrections for the slope and the bias of the developed equations. In total, 517 samples selected based on their spectral variability in 3 countries (Belgium, Ireland, and United Kingdom) from various breeds, cows, and production systems were analyzed by gas chromatography (GC). The samples presenting the largest spectral variability were used to calibrate the prediction of FA by MIR. The remaining samples were used to externally validate the 28 FA equations developed. The 6 methods were (1) partial least squares regression (PLS); (2) PLS + repeatability file (REP); (3) first derivative of spectral data + PLS; (4) first derivative + REP + PLS; (5) second derivative of spectral data + PLS; and (6) second derivative + REP + PLS. Methods were compared on the basis of the cross-validation coefficient of determination (R²cv), the ratio of standard deviation of GC values to the standard error of cross-validation (RPD), and the validation coefficient of determination (R²v). The third and fourth methods had, on average, the highest R²cv, RPD, and R²v. The final equations were built using all GC and the best accuracy was observed for the infrared predictions of C4:0, C6:0, C8:0, C10:0, C12:0, C14:0, C16:0, C18:0, C18:1 trans, C18:1 cis-9, C18:1 cis, and for some groups of FA studied in milk (saturated, monounsaturated, unsaturated, short-chain, medium-chain, and long-chain FA). These equations showed R²cv greater than 0.95. With R²cv equal to 0.85, the MIR prediction of polyunsaturated FA could be used to screen the cow population. As previously published, infrared predictions of FA in fat are less accurate than those developed from FA content in milk (g/dL of milk) and no better results were obtained by using milk FA predictions if no corrections for bias and slope based on reference milk samples with known contents of FA were used. These results indicate the usefulness of equations with R²cv greater than 95% in milk payment systems and the usefulness of equations with R²cv greater than 75% for animal breeding purposes.     

The use of Fourier-transform infrared spectroscopy to predict cheese yield and nutrient recovery or whey loss traits from unprocessed bovine milk samples

Cheese yield is an important technological trait in the dairy industry in many countries. The aim of this study was to evaluate the effectiveness of Fourier-transform infrared (FTIR) spectral analysis of fresh unprocessed milk samples for predicting cheese yield and nutrient recovery traits. A total of 1,264 model cheeses were obtained from 1,500-mL milk samples collected from individual Brown Swiss cows. Individual measurements of 7 new cheese yield-related traits were obtained from the laboratory cheese-making procedure, including the fresh cheese yield, total solid cheese yield, and the water retained in curd, all as a percentage of the processed milk, and nutrient recovery (fat, protein, total solids, and energy) in the curd as a percentage of the same nutrient contained in the milk. All individual milk samples were analyzed using a MilkoScan FT6000 over the spectral range from 5,000 to 900 wavenumber × cm⁻¹. Two spectral acquisitions were carried out for each sample and the results were averaged before data analysis. Different chemometric models were fitted and compared with the aim of improving the accuracy of the calibration equations for predicting these traits. The most accurate predictions were obtained for total solid cheese yield and fresh cheese yield, which exhibited coefficients of determination between the predicted and measured values in cross-validation (1-VR) of 0.95 and 0.83, respectively. A less favorable result was obtained for water retained in curd (1-VR = 0.65). Promising results were obtained for recovered protein (1-VR = 0.81), total solids (1-VR = 0.86), and energy (1-VR = 0.76), whereas recovered fat exhibited a low accuracy (1-VR = 0.41). As FTIR spectroscopy is a rapid, cheap, high-throughput technique that is already used to collect standard milk recording data, these FTIR calibrations for cheese yield and nutrient recovery highlight additional potential applications of the technique in the dairy industry, especially for monitoring cheese-making processes and milk payment systems. In addition, the prediction models can be used to provide breeding organizations with information on new phenotypes for cheese yield and milk nutrient recovery, potentially allowing these traits to be enhanced through selection.     

The use of mid-infrared spectrometry to predict body energy status of Holstein cows

Energy balance, especially in early lactation, is known to be associated with subsequent health and fertility in dairy cows. However, its inclusion in routine management decisions or breeding programs is hindered by the lack of quick, easy, and inexpensive measures of energy balance. The objective of this study was to evaluate the potential of mid-infrared (MIR) analysis of milk, routinely available from all milk samples taken as part of large-scale milk recording and milk payment operations, to predict body energy status and related traits in lactating dairy cows. The body energy status traits investigated included energy balance and body energy content. The related traits of body condition score and energy intake were also considered. Measurements on these traits along with milk MIR spectral data were available on 17 different test days from 268 cows (418 lactations) and were used to develop the prediction equations using partial least squares regression. Predictions were externally validated on different independent subsets of the data and the results averaged. The average accuracy of predicting body energy status from MIR spectral data was as high as 75% when energy balance was measured across lactation. These predictions of body energy status were considerably more accurate than predictions obtained from the sometimes proposed fat-to-protein ratio in milk. It is not known whether the prediction generated from MIR data are a better reflection of the true (unknown) energy status than the actual energy status measures used in this study. However, results indicate that the approach described may be a viable method of predicting individual cow energy status for a large scale of application.     

Milk mid-infrared spectral data as a tool to predict feed intake in lactating Norwegian Red dairy cows

Mid-infrared (MIR) spectroscopy of milk was used to predict dry matter intake (DMI) and net energy intake (NEI) in 160 lactating Norwegian Red dairy cows. A total of 857 observations were used in leave-one-out cross-validation and external validation to develop and validate prediction equations using 5 different models. Predictions were performed using (multiple) linear regression, partial least squares (PLS) regression, or best linear unbiased prediction (BLUP) methods. Linear regression was implemented using just milk yield (MY) or fat, protein, and lactose concentration in milk (Mcont) or using MY together with body weight (BW) as predictors of intake. The PLS and BLUP methods were implemented using just the MIR spectral information or using the MIR together with Mcont, MY, BW, or NEI from concentrate (NEIconc). When using BLUP, the MIR spectral wavelengths were always treated as random effects, whereas Mcont, MY, BW, and NEIconc were considered to be fixed effects. Accuracy of prediction (R) was defined as the correlation between the predicted and observed feed intake test-day records. When using the linear regression method, the greatest R of predicting DMI (0.54) and NEI (0.60) in the external validation was achieved when the model included both MY and BW. When using PLS, the greatest R of predicting DMI (0.54) and NEI (0.65) in the external validation data set was achieved when using both BW and MY as predictors in combination with the MIR spectra. When using BLUP, the greatest R of predicting DMI (0.54) in the external validation was when using MY together with the MIR spectra. The greatest R of predicting NEI (0.65) in the external validation using BLUP was achieved when the model included both BW and MY in combination with the MIR spectra or when the model included both NEIconc and MY in combination with MIR spectra. However, although the linear regression coefficients of actual on predicted values for DMI and NEI were not different from unity when using PLS, they were less than unity for some of the models developed using BLUP. This study shows that MIR spectral data can be used to predict NEI as a measure of feed intake in Norwegian Red dairy cattle and that the accuracy is augmented if additional, often available data are also included in the prediction model.