Among the dairy sector's current concerns, the assessment of global animal health status is a complex challenge. Its multidimensionality means that global monitoring tools are rarely considered. Instead, specific disease detection is often studied separately and, due to financial and ethical issues, uses small-scale data sets focusing on few biomarkers. Several studies have already been conducted using milk Fourier transform mid-infrared (FT-MIR) spectroscopy to detect mastitis and lameness or to quantify health-related biomarkers in milk or blood. Those studies are relevant but they focus mainly on one biomarker or disease. To solve this issue and the small-scale data set, in this study, we proposed a holistic approach using big data obtained from milk recording, including milk yield, somatic cell count, and 27 FT-MIR-based predictors related to milk composition and animal health status. Using 740,454 records collected from 114,536 first-parity Holstein cows in southern Belgium, we performed repeated unsupervised learning algorithms based on Ward's agglomerative hierarchical clustering method to find potential interesting patterns. A divide-and-conquer approach was used to overcome the limitation of computational resources in clustering a relatively large data set. Five groups of records were identified. Differences observed in the fourth group suggested a relationship to metabolic disorders. The fifth group seemed to be related to mastitis. In a second step, we performed a partial least squares discriminant analysis (PLS-DA) to predict the probability of belonging to those specific groups for the entire data set. The obtained global accuracy was 0.77 and the balanced accuracy (i.e., the mean between sensitivity and specificity) of discriminating the fourth and fifth groups was 0.88 and 0.96, respectively. Then, a validation of the interpretation of those groups was performed using 204 milk and blood reference records. The predicted probability associated with the metabolic disorders issue had significant correlations of 0.54 with blood β-hydroxybutyrate, 0.44 with blood nonesterified fatty acids, -0.32 with blood glucose, -0.23 with milk glucose-6-phosphate, and 0.38 with milk isocitrate. In contrast, the predicted probability of belonging to the mastitis group had correlations of 0.69 with milk lactate dehydrogenase, 0.46 with milk N-acetyl-β-d-glucosaminidase, -0.18 with milk free glucose, and 0.16 with milk glucose-6-phosphate. Consequently, these results suggest that the obtained quantitative traits indirectly reflect some of the main health disorders in dairy farming and could be used to monitor dairy cows on a large scale. By using unsupervised learning on large-scale milk recording data and then validating the pattern using reference laboratory measures, we propose a new approach to quickly assess dairy cow health status.

The negative energy balance (NEB) state in dairy cows is a critical factor affecting health, reproduction, and production, particularly during early lactation. Multiple blood and milk biomarkers change when dairy cows are in the NEB state. Direct measurement of NEB is impractical for large-scale use due to costs, necessitating reliance on indirect predictors such as milk mid-infrared (MIR) spectrometry-based predicted biomarkers. However, the genetic relationships between NEB and its potential biomarkers remain unclear. This study aimed to (1) compare measured reference NEB with MIR-predicted NEB (PNEB), a novel energy deficit score (EDS), 15 biomarkers, and 3 production traits; (2) estimate genetic parameters among these traits using a 20-trait repeatability model, quantifying the ability of the 19 other studied traits (logit-transformed EDS (LEDS), 15 biomarkers, and 3 production traits) to genetically predict logit-transformed PNEB (LPNEB); and (3) evaluate the causal effects of LPNEB on the 19 traits through a recursive model. Two datasets were used: dataset I (127 cows, 965 records) provided reference data for objective (1), and dataset II (25,287 first-parity cows, 30,634 records) enabled genetic analysis used for objectives (2) and (3). Traits were analyzed using Pearson correlations, multiple-diagonalization EM-REML-based genetic parameter estimation, and recursive modeling. The studied traits had moderate to moderate-high h² ranging from 0.16 to 0.38. The genetic correlations between LPNEB and the studied traits ranged from -0.60 to 0.85 for LEDS and 0.87 for MIR-predicted blood nonesterified fatty acids (+). Analysis of genetic predictability of LPNEB revealed that the 19 other traits together explained 89% of the genetic variance of LPNEB, with all 15 biomarkers alone contributing the largest fraction with 82%, LEDS alone 65%, nonesterified fatty acids (NEFA) alone 62%, and all traits except LEDS 85%, indicating that LEDS contains useful additional information. Recursive modeling further identified 8 traits, including NEFA and LEDS, as highly dependent on LPNEB, highlighting their potential as robust biomarkers. This study demonstrates the utility of MIR-predicted traits for understanding the genetic mechanisms of NEB and its potential for integration into breeding programs, while emphasizing cautious interpretation of these results due to limitations of MIR-predictions of studied traits to represent directly measured traits.

Summary: The dairy sector is operating in an environment of rapid technological change, driven by advances in big data and artificial intelligence (AI). To remain competitive and sustainable, farms should adopt intelligent, scalable, and privacy-preserving solutions. This paper presents a comprehensive ecosystem that combines multimodal AI, edge computing, and federated learning to enable real-time, data-driven decision making on farms. With emphasis on privacy, local relevance, and research integration, the framework would enable autonomous AI agents to support everyday farm decisions and long-term planning. The graphical abstract depicts a schematic representation of the proposed framework for a future dairy data ecosystem, illustrating interactions among commercial and research farms, edge devices, and the cloud intelligence layer (enterprise and research and development [R&D] clouds). It shows data and model flows, with solid arrows pointing away from farms and dashed arrows pointing toward farms. E = edge layer; FL = federated learning; RAW = raw data.

This article focuses on the creation of a monitoring tool using routinely collected data from milk payment analyses. Milk samples were analyzed through Fourier Transform mid-infrared spectrometry every 1 to 3 days, and their compositions were predicted using machine learning models. Among the predicted parameters, fatty acid profiles appear to be effective indicators of animal status and management practices. In this research, these profiles were summarized using 31 fatty acids or groups of fatty acids. The methodology consists of four steps: hierarchical clustering to detect patterns in a Belgian spectral dataset (N = 774,781), interpretation of the identified seven clusters, development of predictive models applied to a Canadian dataset (N = 670,165), and validation using management information collected from Canadian farms. The identified clusters revealed significant relationships with feeding management strategies and temporal evolutions, highlighting the potential to develop automated alert systems that assist farmers and advisors in herd monitoring.

This research aims to develop a predictive model to discriminate milk produced from a cattle diet either based on grass or not using milk mid-infrared spectrometry and the month of testing (an indirect indicator of the feeding ration). The dataset contained 3,377,715 spectra collected between 2011 and 2021 from 2449 farms and 3 grazing traits defined following the month of testing. Records from 30% of the randomly selected farms were kept in the calibration set, and the remaining records were used to validate the models. Around 90% of the records were correctly discriminated. This accuracy is very good, as some records could be erroneously assigned. The probability of belonging to the GRASS modality allowed confirmation of the model’s ability to detect the transition period even if the model was not trained on this data. Indeed, the probability increased from the spring to the summer and then decreased. The discrimination was mainly explained by the changes in the milk fat, mineral, and protein compositions. A hierarchical clustering from the averaged probability per farm and year highlighted 12 groups illustrating different management practices. The probability of belonging to the GRASS class could be used in a tool counting the number of grazing days.