Using milk mid-infrared spectroscopy to estimate cow-level nitrogen efficiency metrics.

Minimizing pollution from the dairy sector is paramount; one potential cause of such pollution is excess nitrogen. Nitrogen pollution contributes to a deterioration in water quality as well as an increase in both eutrophication and greenhouse gases. It is therefore essential to minimize the loss of nitrogen from the sector, including excretion from the cow. Breeding programs are one potential strategy to improve the efficiency with which nitrogen is used by dairy cows but relies on routine access to individual cow information on how efficiently each cows uses the nitrogen it ingests. A total of 3,497 test-day records for individual cow nitrogen efficiency metrics along with milk yield and the associated milk spectra were used to investigate the ability of milk infrared spectral data to predict these nitrogen traits; both traditional partial least squares regression and neural networks were used in the prediction process. The data originated from 4 farms across 11 years. The nitrogen traits investigated were nitrogen intake, nitrogen use efficiency, and nitrogen balance. Both nitrogen use efficiency and nitrogen balance were calculated considering nitrogen intake, nitrogen in milk, nitrogen in the conceptus, nitrogen used for the growth, nitrogen stored in the reserves, and nitrogen mobilized from the reserves. Irrespective of the nitrogen-related trait being investigated, the best prediction from 4-fold cross-validation were achieved using neural networks that considered both the morning and evening milk spectra along with milk yield, parity, and days in milk in the prediction process. The coefficient of determination in the cross-validation was 0.61, 0.74, and 0.58 for nitrogen intake, nitrogen use efficiency, and nitrogen balance, respectively. In a separate series of validation approaches, the calibration and validation was stratified by herd (n = 4) and separately by year. For these scenarios, partial least squares regression generated more accurate predictions compared with neural networks; the coefficient of determination was always lower than 0.29 and 0.60 when validation was stratified by herd and year, respectively. Therefore, if the variability of the data being predicted in the validation data sets is similar to that in the data used to develop the predictions, then nitrogen-related traits can be predicted with reasonable accuracy. In contrast, where the variability of the data that exists in the validation data set is poorly represented in the calibration data set, then poor predictions will ensue.

Cheese yield and nutrients recovery in the curd predicted by Fourier-transform spectra from individual sheep milk samples.

The objectives of this study were to explore the use of Fourier-transform infrared (FTIR) spectroscopy on individual sheep milk samples for predicting cheese-making traits, and to test the effect of the farm variability on their prediction accuracy. For each of 121 ewes from 4 farms, a laboratory model cheese was produced, and 3 actual cheese yield traits (fresh cheese, cheese solids, and cheese water) and 4 milk nutrient recovery traits (fat, protein, total solids, and energy) in the curd were measured. Calibration equations were developed using a Bayesian approach with 2 different scenarios: (1) a random cross-validation (80% calibration; 20% validation set), and (2) a leave-one-out validation (3 farms used as calibration, and the remaining one as validation set) to assess the accuracy of prediction of samples from external farms, not included in calibration set. The best performance was obtained for predicting the yield and recovery of total solids, justifying for the practical application of the method at sheep population and dairy industry levels. Performances for the remaining traits were lower, but still useful for the monitoring of the milk processing in the case of fresh curd and recovery of energy. Insufficient accuracies were found for the recovery of protein and fat, highlighting the complex nature of the relationships among the milk nutrients and their recovery in the curd. The leave-one-out validation procedure, as expected, showed lower prediction accuracies, as a result of the characteristics of the farming systems, which were different between calibration and validation sets. In this regard, the inclusion of information related to the farm could help to improve the prediction accuracy of these traits. Overall, a large contribution to the prediction of the cheese-making traits came from the areas known as "water" and "fingerprint" regions. These findings suggest that, according to the traits studied, the inclusion of water regions for the development of the prediction equation models is fundamental to maintain a high prediction accuracy. However, further studies are necessary to better understand the role of specific absorbance peaks and their contribution to the prediction of cheese-making traits, to offer reliable tools applicable along the dairy ovine chain.

Prediction of detailed blood metabolic profile using milk infrared spectra and machine learning methods in dairy cattle.

The adoption of preventive management decisions is crucial to dealing with metabolic impairments in dairy cattle. Various serum metabolites are known to be useful indicators of the health status of cows. In this study, we used milk Fourier-transform mid-infrared (FTIR) spectra and various machine learning (ML) algorithms to develop prediction equations for a panel of 29 blood metabolites, including those related to energy metabolism, liver function/hepatic damage, oxidative stress, inflammation/innate immunity, and minerals. For most traits, the data set comprised observations from 1,204 Holstein-Friesian dairy cows belonging to 5 herds. An exception was represented by ß-hydroxybutyrate prediction, which contained observations from 2,701 multibreed cows pertaining to 33 herds. The best predictive model was developed using an automatic ML algorithm that tested various methods, including elastic net, distributed random forest, gradient boosting machine, artificial neural network, and stacking ensemble. These ML predictions were compared with partial least squares regression, the most commonly used method for FTIR prediction of blood traits. Performance of each model was evaluated using 2 cross-validation (CV) scenarios: 5-fold random (CVr) and herd-out (CVh). We also tested the best model's ability to classify values precisely in the 2 extreme tails, namely, the 25th (Q25) and 75th (Q75) percentiles (true-positive prediction scenario). Compared with partial least squares regression, ML algorithms achieved more accurate performance. Specifically, elastic net increased the R2 value from 5% to 75% for CVr and 2% to 139% for CVh, whereas the stacking ensemble increased the R2 value from 4% to 70% for CVr and 4% to 150% for CVh. Considering the best model, with the CVr scenario, good prediction accuracies were obtained for glucose (R2 = 0.81), urea (R2 = 0.73), albumin (R2 = 0.75), total reactive oxygen metabolites (R2 = 0.79), total thiol groups (R2 = 0.76), ceruloplasmin (R2 = 0.74), total proteins (R2 = 0.81), globulins (R2 = 0.87), and Na (R2 = 0.72). Good prediction accuracy in classifying extreme values was achieved for glucose (Q25 = 70.8%, Q75 = 69.9%), albumin (Q25 = 72.3%), total reactive oxygen metabolites (Q25 = 75.1%, Q75 = 74%), thiol groups (Q75 = 70.4%), total proteins (Q25 = 72.4%, Q75 = 77.2.%), globulins (Q25 = 74.8%, Q75 = 81.5%), and haptoglobin (Q75 = 74.4%). In conclusion, our study shows that FTIR spectra can be used to predict blood metabolites with relatively good accuracy, depending on trait, and are a promising tool for large-scale monitoring.

Dry Matter Intake Prediction from Milk Spectra in Sarda Dairy Sheep.

Individual dry matter intake (DMI) is a relevant factor for evaluating feed efficiency in livestock. However, the measurement of this trait on a large scale is difficult and expensive. DMI, as well as other phenotypes, can be predicted from milk spectra. The aim of this work was to predict DMI from the milk spectra of 24 lactating Sarda dairy sheep ewes. Three models (Principal Component Regression, Partial Least Squares Regression, and Stepwise Regression) were iteratively applied to three validation schemes: records, ewes, and days. DMI was moderately correlated with the wavenumbers of the milk spectra: the largest correlations (around ±0.30) were observed at ~1100-1330 cm-1 and ~2800-3000 cm-1. The average correlations between real and predicted DMI were 0.33 (validation on records), 0.32 (validation on ewes), and 0.23 (validation on days). The results of this preliminary study, even if based on a small number of animals, demonstrate that DMI can be routinely estimated from the milk spectra.

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 (CFt)], 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), CFt (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 CFt, 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 CFt, 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.

Identifying Health Status in Grazing Dairy Cows from Milk Mid-Infrared Spectroscopy by Using Machine Learning Methods.

The early detection of health problems in dairy cattle is crucial to reduce economic losses. Mid-infrared (MIR) spectrometry has been used for identifying the composition of cow milk in routine tests. As such, it is a potential tool to detect diseases at an early stage. Partial least squares discriminant analysis (PLS-DA) has been widely applied to identify illness such as lameness by using MIR spectrometry data. However, this method suffers some limitations. In this study, a series of machine learning techniques-random forest, support vector machine, neural network (NN), convolutional neural network and ensemble models-were used to test the feasibility of identifying cow sickness from 1909 milk sample MIR spectra from Holstein-Friesian, Jersey and crossbreed cows under grazing conditions. PLS-DA was also performed to compare the results. The sick cow records had a time window of 21 days before and 7 days after the milk sample was analysed. NN showed a sensitivity of 61.74%, specificity of 97% and positive predicted value (PPV) of nearly 60%. Although the sensitivity of the PLS-DA was slightly higher than NN (65.6%), the specificity and PPV were lower (79.59% and 15.25%, respectively). This indicates that by using NN, it is possible to identify a health problem with a reasonable level of accuracy.

Evaluating the performance of machine learning methods and variable selection methods for predicting difficult-to-measure traits in Holstein dairy cattle using milk infrared spectral data.

Fourier-transform infrared (FTIR) spectroscopy is a powerful high-throughput phenotyping tool for predicting traits that are expensive and difficult to measure in dairy cattle. Calibration equations are often developed using standard methods, such as partial least squares (PLS) regression. Methods that employ penalization, rank-reduction, and variable selection, as well as being able to model the nonlinear relations between phenotype and FTIR, might offer improvements in predictive ability and model robustness. This study aimed to compare the predictive ability of 2 machine learning methods, namely random forest (RF) and gradient boosting machine (GBM), and penalized regression against PLS regression for predicting 3 phenotypes differing in terms of biological meaning and relationships with milk composition (i.e., phenotypes measurable directly and not directly in milk, reflecting different biological processes which can be captured using milk spectra) in Holstein-Friesian cattle under 2 cross-validation scenarios. The data set comprised phenotypic information from 471 Holstein-Friesian cows, and 3 target phenotypes were evaluated: (1) body condition score (BCS), (2) blood ß-hydroxybutyrate (BHB, mmol/L), and (3) κ-casein expressed as a percentage of nitrogen (κ-CN, % N). The data set was split considering 2 cross-validation scenarios: samples-out random in which the population was randomly split into 10-folds (8-folds for training and 1-fold for validation and testing); and herd/date-out in which the population was randomly assigned to training (70% herd), validation (10%), and testing (20% herd) based on the herd and date in which the samples were collected. The random grid search was performed using the training subset for the hyperparameter optimization and the validation set was used for the generalization of prediction error. The trained model was then used to assess the final prediction in the testing subset. The grid search for penalized regression evidenced that the elastic net (EN) was the best regularization with increase in predictive ability of 5%. The performance of PLS (standard model) was compared against 2 machine learning techniques and penalized regression using 2 cross-validation scenarios. Machine learning methods showed a greater predictive ability for BCS (0.63 for GBM and 0.61 for RF), BHB (0.80 for GBM and 0.79 for RF), and κ-CN (0.81 for GBM and 0.80 for RF) in samples-out cross-validation. Considering a herd/date-out cross-validation these values were 0.58 (GBM and RF) for BCS, 0.73 (GBM and RF) for BHB, and 0.77 (GBM and RF) for κ-CN. The GBM model tended to outperform other methods in predictive ability around 4%, 1%, and 7% for EN, RF, and PLS, respectively. The prediction accuracies of the GBM and RF models were similar, and differed statistically from the PLS model in samples-out random cross-validation. Although, machine learning techniques outperformed PLS in herd/date-out cross-validation, no significant differences were observed in terms of predictive ability due to the large standard deviation observed for predictions. Overall, GBM achieved the highest accuracy of FTIR-based prediction of the different phenotypic traits across the cross-validation scenarios. These results indicate that GBM is a promising method for obtaining more accurate FTIR-based predictions for different phenotypes in dairy cattle.

The use of milk Fourier transform mid-infrared spectra and milk yield to estimate heat production as a measure of efficiency of dairy cows.

BACKGROUND: Transformation of feed energy ingested by ruminants into milk is accompanied by energy losses via fecal and urine excretions, fermentation gases and heat. Heat production may differ among dairy cows despite comparable milk yield and body weight. Therefore, heat production can be considered an indicator of metabolic efficiency and directly measured in respiration chambers. The latter is an accurate but time-consuming technique. In contrast, milk Fourier transform mid-infrared (FTIR) spectroscopy is an inexpensive high-throughput method and used to estimate different physiological traits in cows. Thus, this study aimed to develop a heat production prediction model using heat production measurements in respiration chambers, milk FTIR spectra and milk yield measurements from dairy cows. METHODS: Heat production was computed based on the animal's consumed oxygen, and produced carbon dioxide and methane in respiration chambers. Heat production data included 168 24-h-observations from 64 German Holstein and 20 dual-purpose Simmental cows. Animals were milked twice daily at 07:00 and 16:30 h in the respiration chambers. Milk yield was determined to predict heat production using a linear regression. Milk samples were collected from each milking and FTIR spectra were obtained with MilkoScan FT 6000. The average or milk yield-weighted average of the absorption spectra from the morning and afternoon milking were calculated to obtain a computed spectrum. A total of 288 wavenumbers per spectrum and the corresponding milk yield were used to develop the heat production model using partial least squares (PLS) regression. RESULTS: Measured heat production of studied animals ranged between 712 and 1470 kJ/kg BW0.75. The coefficient of determination for the linear regression between milk yield and heat production was 0.46, whereas it was 0.23 for the FTIR spectra-based PLS model. The PLS prediction model using weighted average spectra and milk yield resulted in a cross-validation variance of 57% and a root mean square error of prediction of 86.5 kJ/kg BW0.75. The ratio of performance to deviation (RPD) was 1.56. CONCLUSION: The PLS model using weighted average FTIR spectra and milk yield has higher potential to predict heat production of dairy cows than models applying FTIR spectra or milk yield only.

Usefulness of milk mid-infrared spectroscopy for predicting lameness score in dairy cows.

The objective of this study was to evaluate the ability of milk infrared spectra to predict cow lameness score (LMS) for use as an indicator of cow health on Australian dairy farms, or as an indicator trait for genetic evaluation purposes. The study involved 3,771 cows from 10 farms in Australia. Milk infrared spectra collected during the monthly herd testing were available in all the farms involved in the study. Lameness score was measured once in each herd, within 72 h from a test day, and merged to the closest spectra records. Lameness score was expressed on a scale from 0 to 3, where 0 is assigned to sound cows and scores 1 to 3 are assigned to cows with increased lameness severity. Partial least squares discriminant analysis was used to develop prediction models for classifying sound (score 0) and not-sound cows (i.e., cows walking unevenly, score greater than 0). Discriminant models were tested in a 10-fold random cross-validation process. Milk infrared spectra correctly classified only 57% of the cows walking unevenly and only 59% of the sound cows. When additional predictors (parity, age at calving, days in milk, and milk yield) were included in the prediction model, the model correctly classified 57% of the cows walking unevenly and 62% of the sound cows. The same model applied only to the cows in the first third of lactation correctly classified 66% of the cows walking unevenly and 57% of the sound cows. When the prediction model was used to identify lame cows (scores 2 and 3), only 49% of them were classified as such. These results are considered to be too poor to envisage a practical application of these models in the near future as on-farm tools to provide an indication of LMS. To investigate whether, at this stage, predictions of the LMS could be useful as large-scale phenotypes for animal breeding purposes, we estimated (co)variance components for actual and predicted LMS using 2,670 and 24,560 records, respectively. As the genetic correlation between actual and predicted LMS was not significantly different from zero, predictions of lameness from milk spectra and additional on-farm variables cannot be used, at this stage, as an indicator trait for actual LMS. More research is needed to find better strategies to predict lameness.

Mining data from milk infrared spectroscopy to improve feed intake predictions in lactating dairy cows.

Feed intake is one of the most important components of feed efficiency in dairy systems. However, it is a difficult trait to measure in commercial operations for individual cows. Milk spectrum from mid-infrared spectroscopy has been previously used to predict milk traits, and could be an alternative to predict dry matter intake (DMI). The objectives of this study were (1) to evaluate if milk spectra can improve DMI predictions based only on cow variables; (2) to compare artificial neural network (ANN) and partial least squares (PLS) predictions; and (3) to evaluate if wavelength (WL) selection through Bayesian network (BN) improves prediction quality. Milk samples (n = 1,279) from 308 mid-lactation dairy cows [127 ± 27 d in milk (DIM)] were collected between 2014 and 2016. For each milk spectra time point, DMI (kg/d), body weight (BW, kg), milk yield (MY, kg/d), fat (%), protein (%), lactose (%), and actual DIM were recorded. The DMI was predicted with ANN and PLS using different combinations of explanatory variables. Such combinations, called covariate sets, were as follows: set 1 (MY, BW0.75, DIM, and 361 WL); set 2 [MY, BW0.75, DIM, and 33 WL (WL selected by BN)]; set 3 (MY, BW0.75, DIM, and fat, protein, and lactose concentrations); set 4 (MY, BW0.75, DIM, 33 WL, fat, protein, and lactose); set 5 (MY, BW0.75, DIM, 33 WL, and visit duration in the feed bunk); set 6 (MY, DIM, and 33 WL); set 7 (MY, BW0.75, and DIM); set-WL (included 361 WL); and set-BN (included just 33 selected WL). All models (i.e., each combination of covariate set and fitting approach, ANN or PLS) were validated with an external data set. The use of ANN improved the performance of models 2, 5, 6, and BN. The use of BN combined with ANN yielded the highest accuracy and precision. The addition of individual WL compared with milk components (set 2 vs. set 3) did not improve prediction quality when using PLS. However, when ANN was employed, the model prediction with the inclusion of 33 WL was improved over the model containing only milk components (set 2 vs. set 3; concordance correlation coefficient = 0.80 vs. 0.72; coefficient of determination = 0.67 vs. 0.53; root mean square error of prediction 2.36 vs. 2.81 kg/d). The use of ANN and the inclusion of a behavior parameter, set 5, resulted in the best predictions compared with all other models (coefficient of determination = 0.70, concordance correlation coefficient = 0.83, root mean square error of prediction = 2.15 kg/d). The addition of milk spectra information to models containing cow variables improved the accuracy and precision of DMI predictions in lactating dairy cows when ANN was used. The use of BN to select more informative WL improved the model prediction when combined with cow variables, with further improvement when combined with ANN.

An attempt at predicting blood ß-hydroxybutyrate from Fourier-transform mid-infrared spectra of milk using multivariate mixed models in Polish dairy cattle.

Fourier transform mid-infrared (FT-MIR) spectra of milk are commonly used for phenotyping of traits of interest through links developed between the traits and milk FT-MIR spectra. Predicted traits are then used in genetic analysis for ultimate phenotypic prediction using a single-trait mixed model that account for cows' circumstances at a given test day. Here, this approach is referred to as indirect prediction (IP). Alternatively, FT-MIR spectral variable can be kept multivariate in the form of factor scores in REML and BLUP analyses. These BLUP predictions, including phenotype (predicted factor scores), were converted to single-trait through calibration outputs; this method is referred to as direct prediction (DP). The main aim of this study was to verify whether mixed modeling of milk spectra in the form of factors scores (DP) gives better prediction of blood ß-hydroxybutyrate (BHB) than the univariate approach (IP). Models to predict blood BHB from milk spectra were also developed. Two data sets that contained milk FT-MIR spectra and other information on Polish dairy cattle were used in this study. Data set 1 (n = 826) also contained BHB measured in blood samples, whereas data set 2 (n = 158,028) did not contain measured blood values. Part of data set 1 was used to calibrate a prediction model (n = 496) and the remaining part of data set 1 (n = 330) was used to validate the calibration models, as well as to evaluate the DP and IP approaches. Dimensions of FT-MIR spectra in data set 2 were reduced either into 5 or 10 factor scores (DP) or into a single trait (IP) with calibration outputs. The REML estimates for these factor scores were found using WOMBAT. The BLUP values and predicted BHB for observations in the validation set were computed using the REML estimates. Blood BHB predicted from milk FT-MIR spectra by both approaches were regressed on reference blood BHB that had not been used in the model development. Coefficients of determination in cross-validation for untransformed blood BHB were from 0.21 to 0.32, whereas that for the log-transformed BHB were from 0.31 to 0.38. The corresponding estimates in validation were from 0.29 to 0.37 and 0.21 to 0.43, respectively, for untransformed and logarithmic BHB. Contrary to expectation, slightly better predictions of BHB were found when univariate variance structure was used (IP) than when multivariate covariance structures were used (DP). Conclusive remarks on the importance of keeping spectral data in multivariate form for prediction of phenotypes may be found in data sets where the trait of interest has strong relationships with spectral variables.