Adulteration of higher priced milks with cheaper ones to obtain extra profit can adversely affect consumer health and the market. In this study, pure buffalo milk (BM), goat milk (GM), camel milk (CM), and their mixtures with 5–50% (vol/vol) cow milk or water were used. Mid-infrared spectroscopy (MIRS) combined with modern statistical machine learning was used for the discrimination and quantification of cow milk or water adulteration in BM, GM, and CM. Compared to partial least squares (PLS), modern statistical machine learning—especially support vector machines (SVM), projection pursuit regression (PPR), and Bayesian regularized neural networks (BRNN)—exhibited superior performance for the detection of adulteration. The best prediction models for the different predictive traits are as follows: The binary classification models developed by SVM resulted in differentiation of CM-cow milk, and GM/CM-water mixtures. PLS resulted in differentiation of BM/GM-cow milk and BM-water mixtures. All of the above models have 100% classification accuracy. SVM was used to develop multi-classification models for identifying the high and low proportions of cow milk in BM, GM, and CM, as well as the high and low proportions of water adulteration in BM and GM, with correct classification rates of 94%, 100%, 100%, 99%, and 100%, respectively. In addition, a PLS-based model was developed for identifying the high and low proportions of water adulteration in CM, with correct classification rates of 100%. A regression model for quantifying cow milk in BM was developed using PCA + BRNN, with RMSEV = 5.42%, and R_V² = 0.88. A regression model for quantifying water adulteration in BM was developed using PCA + PPR, with RMSEV = 1.70%, and R_V² = 0.99. Modern statistical machine learning improved the accuracy of MIRS in predicting BM, GM, and CM adulteration more effectively than PLS.

Mid-infrared spectroscopy (MIRS) is increasingly used as a rapid and effective analytical method for the quantitative prediction of detailed milk composition, such as minerals, fatty acids, and AA. These analyses require the transportation of samples to a certified laboratory. In this case, storage time may affect MIRS and its prediction results. This study aimed to determine the effect of milk storage time on MIRS and its predictions of AA content. A total of 373 individual milk samples for the development of AA content prediction equations were collected from 7 commercial dairy farms (dataset 1), and 103 individual milk samples for the analysis of the effect of storage time were collected from 2 farms (dataset 2). First, separate quantitative prediction models based on dataset 1 were developed for each AA using partial least squares regression; the accuracy of prediction was assessed using a cross-validation set. Second, repeatability and reproducibility of the predictions of AA were calculated using milk samples in dataset 2, whose spectra were measured once a day for 7 consecutive days after sampling storing at 4°C, to assess the effect of storage time on the consistency of MIRS predicted AA content. Moderate to high prediction accuracy of AA was achieved, with the ratio of performance to deviation of the cross-validation set and the R² of the cross-validation set being in the range of 1.59 (Gly) to 2.39 (Leu), and from 0.58 (Gly) to 0.79 (Leu), respectively. Results demonstrated that the absorbance of spectral points was affected by milk storage time, especially in the absorption areas associated with fat, protein, lactose, SCC, urea, and acetone. However, the predictions of all the AA by MIRS were repeatable and reproducible across different milk storage times until 6 d after sampling, except for Tyr, His, and Phe, with repeatability and reproducibility both greater than or close to 90%. In conclusion, for milk stored at 4°C and preserved with bronopol, MIRS can provide relatively consistent and accurate predictions for AA content for 0 to 6 d after milk collection. However, a shorter storage time, such as within 3 d after collection, is recommended when conditions permit.

Accurate identification of cows' likelihood of conception during the period from recent calving to the first artificial insemination (AI) will provide assistance in managing the fertility of dairy cows and contribute to the economic prosperity and sustainability of the farm. The purpose of this study was to use FTIR spectroscopy collected from recent calving to the first artificial insemination (AI) to predict the cow's likelihood of conception to first AI, first or second AI. This study specifically focused on the role of FTIR spectral and farm data collected at different time windows in improving the accuracy of model for predicting the cow's likelihood of conception to first AI, first or second AI. From 2019 to 2023, fertility information of 10,873 Holstein dairy cows in China were collected, coupled with 21,928 spectral data. First, cows were classified into "good" and "poor." Strategy 1 (S1) defined "good" as cows conceived at first AI and "poor" as the others. Strategy 2 (S2) defined "good" as cows conceived at first or second AI and "poor" as the others. Second, the partial least squares discriminant analysis was used to establish models for predicting the likelihood of conception to first AI, first or second AI. The model was assessed using a cross-validation (CV) set and herd-independent-external validation (HEV) set. The study also focused on examining the potential correlation between the accuracy of prediction and the period of spectral and farm data collection by analyzing the diagnostic performance of the model at 8 different time windows: '0 to 7 d postpartum (dpp)', '8 to 14 dpp', '15 to 21 dpp', '22 to 30 dpp', '31 to 45 dpp', '46 to 60 dpp', ' ≥ 61 dpp', and '0 to 7 d before the first AI'. The results showed that the model based on S1 performed better while in proximity to the first AI with AUC_CV and AUC_HEV of 0.621 and 0.633, respectively. Models based on S2 exhibited superior performance throughout the late phase of uterine involution. The optimal model was developed by using spectral data collected from '22 to 30 dpp'. The AUC_CV and AUC_HEV were 0.644 and 0.660, respectively, which was higher than that of S1. This study demonstrates the potential of using FTIR spectral data to predict the cow's ability to conceive. The model developed from data collected within a certain time window exhibited better prediction accuracy, particularly during 22 to 30 dpp and 0 to 7 d before the first AI. This study offers novel perspectives on alternate approaches for assessing the fertility of cows, which will contribute to the regularization and sustainability of farms, as well as to the precision management of agriculture.

Mastitis significantly impacts both the yield and quality of milk. The somatic cell count (SCC) and differential somatic cell count (DSCC), which are related to immune cells, are primary indicators for assessing mammary gland health. In this study, eight previously established mid-infrared spectroscopy models were utilized to predict the content of milk protein fractions (αs1-CN, β-CN, κ-CN, total CN, α-LA, β-LG, IgG, and LF) in milk samples from 21,388 lactating cows across 33 herds. Four linear mixed models were applied to analyze the secretion patterns of milk protein fractions by days in milk (DIM) and parity, their variations under different mastitis conditions, and their associations with the somatic cell score (SCS), DSCC, and immune cell counts (PMN + LYM score (PMN + LYMS) and MAC score (MACS)). The primary findings of the investigation comprised the following: (1) IgG was higher in early lactation, decreased with advancing lactation days, and slightly increased in late lactation, while seven other protein factions decreased from early to peak lactation and increased during mid-to-late lactation. Parity influenced all milk protein fractions except αs1-CN, with total CN, β-CN, and α-LA decreasing and κ-CN, β-LG, IgG, and LF increasing as parity increased (p < 0.05). (2) Mastitis significantly reduced the milk yield, fat percentage, protein percentage, and the contents of total CN, β-CN, κ-CN, and α-LA while increasing β-LG, IgG, and LF. (3) The SCS was negatively correlated with milk yield and α-LA but positively correlated with the fat percentage, protein percentage, κ-CN, β-LG, IgG, and LF. (4) When the DSCC increased to 50%, the milk yield decreased, while the milk protein percentage and κ-CN content significantly increased (p < 0.05). When the DSCC exceeded 50%, the fat percentage, protein percentage, total casein, αs1-CN, β-CN, κ-CN, β-LG, IgG, and LF decreased, while the α-LA content increased (p < 0.05). (5) When the PMN + LYMS increased, the milk yield and α-LA content rose, while the milk fat percentage, the milk protein percentage, and the contents of αs1-CN, β-CN, κ-CN, total CN, β-LG, IgG, and LF decreased (p < 0.05). Conversely, when the MACS increased, the milk yield and α-LA content declined, whereas the milk fat percentage, the milk protein percentage, and the contents of αs1-CN, β-CN, κ-CN, total CN, β-LG, IgG, and LF increased (p < 0.05). This study offers valuable insights into enhancing milk product quality, advancing the early diagnosis and mechanistic research of bovine mastitis, and the sustainable development of the dairy farming industry.

The potential of milk Fourier-transform mid-infrared (FTIR) spectroscopy in predicting cow fertility has been extensively examined, but largely based on the number of services per conception (NSC). Compared with NSC, Calving-to-conception interval (CCI) may be a more critical factor influencing the profitability, productivity, and sustainability of dairy herds, as it reflects days to first breeding, voluntary waiting period, NSC, and service intervals. Our objectives were to evaluate the ability of FTIR spectroscopy and farm data collected from early lactation to predict CCI length postcalving in Holstein cows from a highly productive TMR system. We also sought to identify the most informative milk sampling periods for CCI prediction. From January 2019 to December 2023, FTIR spectra records, cow information, milk recording information, and fertility information were collected from 28,434 Holstein cows within 13 dairy farms in China. First, cows were classified into long calving-to-conception interval (LCCI) and short calving-to-conception interval (SCCI) groups based on 2 strategies. Strategy 1 defined LCCI as cows with a CCI longer than 150 d and SCCI as cows with a CCI shorter than 150 d. Strategy 2 employed a similar method but with a CCI threshold of 90 d. Second, partial least squares discriminant analysis was employed to develop prediction models for the classification of LCCI and SCCI cows. The performance of models was assessed using herd-independent cross-validation. These analyses were conducted separately for the complete dataset as well as for each of the 9 subsets stratified based on postcalving time windows. The results showed that the area under the receiver operating characteristic curve of cross-validation (AUC_CV) varied from 0.461 to 0.675 across different predictors, strategies, and time windows. Across all strategies, prediction accuracy was highest for models developed using data from time windows 22 to 30 days postpartum (dpp) and >60 dpp. The classification model, developed using standard normal variate preprocessed spectra, cow information, milk yield, SCS, and fat-to-protein ratio (FPR) data collected from the time window 22 to 30 dpp, demonstrated the best performance in strategy 1. The values of AUC_CV, sensitivity of cross-validation (SENS_CV), and specificity of cross-validation (SPEC_CV) were 0.650, 0.519, and 0.706, respectively. The best-performing classification model based on strategy 2 was developed using Savitzky-Golay preprocessed spectra, cow information, milk yield, SCS, and FPR data collected from the time window >60 dpp, with AUC_CV, SENS_CV, and SPEC_CV values of 0.675, 0.552, and 0.712, respectively. In conclusion, FTIR and farm data collected from early lactation could distinguish between cows with different CCI lengths with moderate accuracy. The time window 22 to 30 dpp could provide more effective and accurate predictions for the future fertility of dairy cows, and its use and implementation should be considered in practical farm production. Our study highlights the future application of high-throughput phenotyping technologies in precision livestock farming and offers novel insights into alternate methods for assessing cow fertility.

A robust model of buffalo milk based on Fourier Transform Mid-Infrared Spectroscopy (FT-MIRS) is lacking and is difficult to complete quickly. Therefore, this study used 614 milk samples from two buffalo farms from south and central China for FT-MIRS to explore the potential of predicting buffalo milk fat, milk protein, and total solids (TS), providing a rapid detection technology for the determination of buffalo milk composition content. It also explored the rapid transformation and application of the model in spatio-temporal dimensions, providing reference strategies for the rapid application of new models and for the establishment of robust models. Thus, a large number of phenotype data can be provided for buffalo production management and genetic breeding. In this study, models were established by using 12 pre-processing methods, artificial feature selection methods, and partial least squares regression. Among them, a fat model with PLSR + SG (w = 15, p = 4) + 302 wave points, a protein model with PLSR + SG (w = 7, p = 4) + 333 wave points, and a TS model with PLSR + None + 522 wave points had the optimal prediction performance. Then, the TS model was used to explore the application strategies. In temporal dimensions, the TS model effectively predicted the samples collected in a contemporaneous period (RPD_V (Relative Analytical Error of Validation Set) = 3.45). In the spatial dimension, at first, the modeling was conducted using the samples from one farm, and afterward, 30–70% of a sample from another farm was added to the debugging model. Then, we found that the predictive ability of the samples from the other farm gradually increased. Therefore, it is possible to predict the composition of buffalo milk based on FT-MIRS. Moreover, when using the two application strategies that predicted contemporaneous samples as the model, and adding 30–70% of the samples from the predicted farm, the model application effect can be improved before the robust model has been fully developed.