Cow-level evaluation of plasma essential amino acid clusters and their association with lactating dairy cow...

Studying changes in plasma EAA concentration and profile may provide a foundation for precision protein nutrition strategies for feeding dairy cows. Using data from a systematic review, Letelier et al. (2022a) identified 2 EAA concentration and profile clusters that differed in performance (DMI, daily milk energy output, fat yield and protein yield) and measures of feed and N use efficiency. However, controlled trials are necessary to evaluate these findings with individual cow-level data. We hypothesized that the EAA concentration and profile at the group level obtained from the literature would result in comparable performance and efficiency measures when evaluated at the individual level. Hence, this study aimed to evaluate the clusters identified in the review of Letelier et al. (2022a) using a supervised machine learning model to determine whether the differences in performance and efficiencies between clusters were consistent when assessed with an independent empirical dataset. These data were from a study of 62 cows stratified in 4 lactation stages (early, mid-early, mid-late, and late) fed diets of 13.6%, 15.2%, 16.7%, or 18.3% CP (DM basis) for 8 wk. A random forest algorithm was trained for cluster identification using data from the review published by Letelier et al. (2022a), with Arg, His, Ile, Leu, Lys, Met, Phe, Thr, and Val concentration and profile (i.e., the molar proportion in total plasma EAA concentration) as features and clusters as outputs. A leave-one-trial-out cross-validation approach evaluated the model's classification ability, achieving accuracy, precision, recall, and F₁ score of 84.4%, 87.3%, 85.7%, and 86.5% for cluster 1, and 84.4%, 80.5%, 82.5%, and 81.5% for cluster 2, respectively. The model was then retrained using the entire 62-cow dataset. The EAA concentration and profile were assessed in wk 4 and 8, where 33 and 38 cows were classified as cluster 1, and 29 and 24 as cluster 2, respectively. Cows in cluster 2 exhibited greater concentrations of Arg (11.8%), His (24.5%), Ile (18.4%), Leu (38.8%), Lys (16%), Met (14.4%), Phe (15.2%), Trp (4.5%), and Val (26.3%) than cows in cluster 1. In addition, cluster 2 cows were determined to have 2.53% greater DMI (28.4 ± 0.23 vs. 27.7 ± 0.22 kg/d) and 2.34% greater ECM yield (42.8 ± 0.51 vs. 43.8 ± 0.54 kg/d), but 2.81% lower N use efficiency (100 × milk true protein N/N intake; 28.5% ± 0.42% vs. 27.7% ± 0.44%) than cows in cluster 1. The was no difference in feed efficiency and milk component concentration between clusters. These findings confirm that the group-level responses previously observed between EAA-based clusters and measures of dairy cow performance also exist at the individual cow level. However, research with larger datasets is needed to clarify the nature of the relationship between EAA-based clusters and measures of performance, feed efficiency, and N use efficiency.