Unsupervised clustering identifies candidate metabolic profiles, including an acidotic profile detectable before calving, in periparturient dairy cows
4 affiliations
- 1Ningxia Hui Autonomous Region Animal Husbandry Workstation, Yinchuan, 750002 China
- 2Ningxia State Farm Dairy Co., Ltd., Yinchuan, 750200 China
- 3Yinchuan Animal Husbandry Technology Extension and Service Center, Yinchuan, 750000 China
- 4Institute of Animal Science and Veterinary Medicine, Shanghai Academy of Agricultural Sciences, Minhang, 201106 China
Abstract
The metabolic status of periparturient dairy cows is highly heterogeneous, making it difficult for traditional assessment methods based on a single biomarker to comprehensively reflect systemic metabolic dysregulation. This study aimed to apply unsupervised clustering to identify distinct metabolic phenotypes using integrated blood physiological indicators and to evaluate the prepartum identifiability of metabolically compromised profiles. Venous blood samples were collected from 316 healthy Holstein cows (from − 7 to + 7 days relative to calving) across seven commercial farms in China for blood gas and electrolyte analysis. After data cleaning and variable selection, six key indicators—pH, PCO₂, HCO₃⁻, glucose, potassium, and sodium—were standardized and subjected to K-means clustering. The optimal number of clusters was determined to be three, yielding the following phenotypes: (1) a “metabolically compromised” group characterized by acidosis (low pH), hypercapnia (elevated PCO₂), hypoglycemia, and electrolyte imbalance; (2) a “normoacidemic–hyperkalemic” group with normal acid–base status but persistently elevated potassium; and (3) a “metabolically homeostatic” group exhibiting alkalotic tendencies, higher sodium and glucose levels, and stable electrolyte balance. A substantial proportion of cows already belonged to the metabolically compromised phenotype before calving, and its prevalence peaked during the mid-postpartum period (+ 4 to + 7 d). In unadjusted analyses, cows in this cluster exhibited modestly elevated postpartum β-hydroxybutyrate concentrations (P = 0.001) and significantly reduced milk protein percentage (P < 0.001); however, the BHBA association was not significant after adjusting for farm of origin in a mixed-effects model (P = 0.622), whereas the milk protein association remained robust. These findings indicate that impaired metabolic adaptation in this cluster is more consistently reflected by reduced milk protein than by elevated BHBA. Based on cluster centroids and validated by exploratory ROC analysis, prepartum thresholds for identifying the metabolically compromised profile were proposed: venous pH < 7.25 (AUC = 0.928), PCO₂ > 58.75 mmHg (AUC = 0.956), and glucose < 27.5 mg/dL (AUC = 0.876). Potassium was not retained as a discriminatory threshold because its AUC was only 0.577. These thresholds are exploratory and require validation in independent cohorts. These findings demonstrate that periparturient dairy cows segregate into three distinct metabolic states, and that a metabolically compromised acidotic profile can be identified antepartum. This provides a foundation for early risk stratification; future intervention studies are needed to determine whether targeting this profile improves clinical outcomes.
Related
- Decoding candidemia in critically ill patients: unsupervised clustering identifies three unique phenotypes
- Concurrent Supervised-Unsupervised Representative Subspace Clustering (CSU-RSC)
- Unsupervised Clustering Enables Cohort-Specific Variant Recalibration in Crops
- Clinical Heterogeneity in Myotonic Dystrophy Type 1: Identifying Phenotype Categories Using Unsupervised Clustering.
- Unsupervised source separation and clustering of soundscape recordings predicts avian richness in hyperdiverse systems
BibTeX
@article{tian2026unsupervised,
title = {{Unsupervised clustering identifies candidate metabolic profiles, including an acidotic profile detectable before calving, in periparturient dairy cows}},
author = {Jia Tian and Yanmei Zhang and Li’e Hou and Yongjun Liu and Jiamin Zhou and Kun Wang and Miaomiao Ma and Tingting Lu and Ning Ma and Yafei Chen and Xin Li and Yu Wang},
journal = {Tropical Animal Health and Production},
year = {2026},
month = nov,
doi = {10.1007/s11250-026-05438-3},
url = {https://doi.org/10.1007/s11250-026-05438-3},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Unsupervised clustering identifies candidate metabolic profiles, including an acidotic profile detectable before calving, in periparturient dairy cows
AU - Jia Tian
AU - Yanmei Zhang
AU - Li’e Hou
AU - Yongjun Liu
AU - Jiamin Zhou
AU - Kun Wang
AU - Miaomiao Ma
AU - Tingting Lu
AU - Ning Ma
AU - Yafei Chen
AU - Xin Li
AU - Yu Wang
PY - 2026
DA - 2026/11/04
JO - Tropical Animal Health and Production
DO - 10.1007/s11250-026-05438-3
UR - https://doi.org/10.1007/s11250-026-05438-3
AB - The metabolic status of periparturient dairy cows is highly heterogeneous, making it difficult for traditional assessment methods based on a single biomarker to comprehensively reflect systemic metabolic dysregulation. This study aimed to apply unsupervised clustering to identify distinct metabolic phenotypes using integrated blood physiological indicators and to evaluate the prepartum identifiability of metabolically compromised profiles. Venous blood samples were collected from 316 healthy Holstein cows (from − 7 to + 7 days relative to calving) across seven commercial farms in China for blood gas and electrolyte analysis. After data cleaning and variable selection, six key indicators—pH, PCO₂, HCO₃⁻, glucose, potassium, and sodium—were standardized and subjected to K-means clustering. The optimal number of clusters was determined to be three, yielding the following phenotypes: (1) a “metabolically compromised” group characterized by acidosis (low pH), hypercapnia (elevated PCO₂), hypoglycemia, and electrolyte imbalance; (2) a “normoacidemic–hyperkalemic” group with normal acid–base status but persistently elevated potassium; and (3) a “metabolically homeostatic” group exhibiting alkalotic tendencies, higher sodium and glucose levels, and stable electrolyte balance. A substantial proportion of cows already belonged to the metabolically compromised phenotype before calving, and its prevalence peaked during the mid-postpartum period (+ 4 to + 7 d). In unadjusted analyses, cows in this cluster exhibited modestly elevated postpartum β-hydroxybutyrate concentrations (P = 0.001) and significantly reduced milk protein percentage (P < 0.001); however, the BHBA association was not significant after adjusting for farm of origin in a mixed-effects model (P = 0.622), whereas the milk protein association remained robust. These findings indicate that impaired metabolic adaptation in this cluster is more consistently reflected by reduced milk protein than by elevated BHBA. Based on cluster centroids and validated by exploratory ROC analysis, prepartum thresholds for identifying the metabolically compromised profile were proposed: venous pH < 7.25 (AUC = 0.928), PCO₂ > 58.75 mmHg (AUC = 0.956), and glucose < 27.5 mg/dL (AUC = 0.876). Potassium was not retained as a discriminatory threshold because its AUC was only 0.577. These thresholds are exploratory and require validation in independent cohorts. These findings demonstrate that periparturient dairy cows segregate into three distinct metabolic states, and that a metabolically compromised acidotic profile can be identified antepartum. This provides a foundation for early risk stratification; future intervention studies are needed to determine whether targeting this profile improves clinical outcomes.
SN - 0049-4747
SN - 1573-7438
ER -