Source-linked AI summary
Meta-clustering of milk mid-infrared spectra identifies dairy cow groups associated with negative energy balance in early lactation
T. Touil, E. R. Paquet
TL;DR
Monitoring cow health and performance is challenging and costly for dairy producers. The study identified five milk-spectrum meta-clusters, with significant associations to production traits and indications of negative energy balance.
Problem
Monitoring cow health and performance is challenging and incurs substantial costs for dairy producers.
Method
The study used similarity among clustering results to identify meta-clusters and considered the approach that recapitulated most meta-clusters most appropriate.
Results
Five meta-clusters were identified, significantly associated with production traits; meta-clusters 1, 2, and 4 were likely experiencing negative energy balance.
Takeaways & Limitations
The identified meta-clusters provide groups associated with production traits and differing negative-energy-balance states.
Takeaways & Limitations
Only milk-derived traits were available; direct blood biomarkers of negative energy balance were not included.
Abstract
from arXiv · showhide
Clustering methods have been used to identify distinct groups of milk samples, cows, or herds. Fourier-transform infrared (FTIR) spectroscopy, particularly mid-infrared (MIR) spectroscopy, has been applied to individual cow milk samples to predict various milk traits. Applying clustering directly to MIR spectral data may reveal latent groups of cows associated with milk traits or health disorders and can help prevent these conditions or monitor at-risk animals. This study aimed to identify groups of individual dairy cows in early lactation directly from milk MIR spectra and to analyze their associations with milk traits. Using a dataset of 407,632 individual milk MIR records from 3,408 commercial farms, we combined (i) spectral filtering that selects informative wavenumbers, (ii) two dimensionality-reduction methods: principal component analysis (PCA) and an autoencoder, and (iii) two clustering algorithms: k-means and spectral clustering to yield eight different clustering approaches. We regrouped the assigned clusters into meta-clusters that encompassed the most similar ones identified by the eight approaches. Our results revealed five distinct meta-clusters of early-lactation individual dairy cows significantly associated with milk traits. Despite substantial differences, the eight approaches converged on the same five meta-clusters, and the classic, computationally efficient PCA-based k-means approach using the full spectrum recaptured clusters identified by more sophisticated, computationally intensive approaches. The five meta-clusters were strongly associated with DIM and appeared to reflect a gradient of negative energy balance (NEB) severity: severe, moderate, and possibly mild, while the remaining two likely represented cows recovering from NEB, one with rapid restoration of energy balance and one in early recovery.
Introduction
Monitoring dairy cow health and performance is challenging and costly, while existing clustering studies had not identified cow groups directly from MIR spectra collected from many commercial-farm cows. This study therefore compared direct MIR-based clustering approaches and examined their agreement and associations with milk traits.
- Monitoring cow health and performance is challenging and incurs substantial costs for dairy producers.
- Previous studies grouped cows or herds using blood biomarkers, predicted biomarkers, milk biomarkers, fatty acids, or atypical MIR spectra.
- None of these studies identified clusters directly from MIR spectral data collected from many individual cows on commercial farms.
- MIR spectra may reveal latent cow groups associated with distinct milk-trait profiles and different metabolic states.
- The study compared direct MIR clustering approaches, assessed agreement among their cluster structures, and analyzed cluster associations with milk traits in thousands of early-lactation cows.
Materials and methods
The study assembled and standardized a large MIR dataset, reduced spectral dimensionality, and applied eight combinations of filtering, representation, and clustering. Cluster assignments were then compared and related to milk traits and days in milk in a restricted early-lactation dataset.
- Data collection and preprocessing: 407,632 milk samples from 291,034 cows on 3,408 Québec commercial farms formed the initial MIR dataset.
- Data collection and preprocessing: One randomly selected spectrum per cow and spectra from three major instruments were retained, yielding 214,869 spectra focused on inter-cow differences.
- Dimensionality reduction and clustering: The workflow combined full-spectrum or 212-wavenumber filtering, PCA or autoencoder reduction, and k-means or spectral clustering into eight approaches.
- Dimensionality reduction and clustering: PCA retained 20 components explaining 95.15% of spectral variability for the unfiltered data and three components explaining 96.51% after F16 filtering.
- Dimensionality reduction and clustering: K-means and spectral clustering used k = 4; spectral clustering required random subsampling because of its memory demands, followed by k-nearest-neighbors label propagation for the full dataset.
Results and discussion
Eight clustering approaches converged on five meta-clusters of early-lactation cows, with the deterministic PCA-based k-means approach recapturing most clusters. These meta-clusters were associated with DIM and milk-trait profiles consistent with differing NEB severity and recovery stages.
- Meta-Clusters Interpretation: Five meta-clusters were identified from similarities among assignments produced by the eight clustering approaches.The meta-clusters represented groups of assignments that overlapped most frequently.
- Meta-Clusters Interpretation: The PCA-based k-means approach identified most meta-clusters and was considered the most appropriate method because it was deterministic and efficient.It recapitulated meta-clusters shared across multiple approaches.
- Interpreting Associations of Meta-Clusters with DIM: Meta-clusters differed in size, ranging from 2,639 cows in meta-cluster 4 to 60,577 in meta-cluster 5, and DIM and milk-production traits were significantly associated with them.Meta-cluster 4 comprised 1.49% of cows, whereas meta-cluster 5 comprised 34.16%.
- Interpreting Associations of Meta-Clusters with Milk Traits: Meta-clusters 1, 2, and 4 were likely experiencing NEB, with milk-fat, fatty-acid, BHB, lactose, and protein patterns indicating moderate to severe metabolic differences.Meta-cluster 4 showed high preformed fatty acids, high BHB, and strong lactose depression, consistent with severe NEB and possibly subclinical ketosis.
- Interpreting Associations of Meta-Clusters with DIM: Meta-clusters 3 and 5 likely represented recovery from NEB, with meta-cluster 3 suggesting rapidly resolved mild NEB and meta-cluster 5 early recovery.Meta-cluster 3 had the lowest fat and protein contents, while meta-cluster 5 showed only mild deviations from the average profile.
- Limitations: The study’s interpretation was limited by the absence of direct blood biomarkers and additional cow measurements, and by restricting the analysis to early lactation.Blood NEFA, BHB, and glucose could have confirmed associations with NEB levels; DMI, BCS, and BW could have strengthened interpretation.
Conclusion
The study identified five early-lactation dairy-cow meta-clusters from MIR spectra using eight clustering approaches, with the groups strongly associated with DIM and milk traits. The PCA-based k-means method using all wavenumbers recovered most meta-clusters, while the clusters appeared to represent differing NEB severity and recovery states.
- Five meta-clusters of individual dairy cows were identified from MIR data using eight approaches combining spectral filtering, dimensionality reduction, and clustering.
- The PCA-based k-means approach using all wavenumbers identified most meta-clusters, including clusters shared across methods and a rarer cluster detected by few approaches.The authors described this method as deterministic and efficient.
- All five meta-clusters were strongly associated with DIM, and three represented cows likely experiencing NEB with severe, moderate, or possibly mild severity.
- The remaining two meta-clusters likely represented cows recovering from NEB, including one with rapidly resolved mild NEB and one in early recovery.
- The NEB interpretations were less precise than interpretations based on blood metabolites and require validation in future studies.
- The approach may support dairy producers in herd monitoring and decision-making related to cow health and performance.
the writing process
The authors used ChatGPT to improve the manuscript’s English and reported that all authors reviewed, verified, and edited the final version.
- ChatGPT was used to improve the English, and all authors reviewed, verified, and edited the final manuscript version.