Abstract #T149

# T149
Milk composition of Holstein cows through two lactations.
Alex Pape*1, Heather M. Dann1, David M. Barbano2, Richard J. Grant1, 1William H. Miner Agricultural Research Institute, Chazy, NY, 2Department of Food Science, Northeast Dairy Food Research Center, Cornell University, Ithaca, NY.

The objective of this study was to measure how certain milk composition features vary with parity by using them to build logistic regression models to differentiate between lactations 1 and 2. Understanding differences in milk composition between lactations is useful for herd diagnostics and troubleshooting. The data came from 50 Holstein cows for which a complete set of milk samples from both lactations was available. Sampling for a lactation was considered complete if samples from at least 20 d (3 samples/d weighted proportionally by milk yield) distributed over at least 8 mo were available. The data set consisted of 2,783 observations. Milk samples were analyzed by mid-infrared methodology. Composition features were normalized relative to the weekly mean and standard deviation (SD) for the herd to eliminate the effect of systematic change over time. Data were stratified into 4 stages of lactation: fresh (0–21 DIM), early (22–90 DIM), mid (91–150 DIM), and late (>150 DIM). Three types of models were used: a collective model with principal components of all composition features, a collective model with selected variables, and individual models for selected variables. The selected variables were fat (4.00 ± 0.65%; mean ± SD), true protein (3.08 ± 0.31%), anhydrous lactose (4.62 ± 0.20%), de novo fatty acids (FA; 23.3 ± 2.3 g/100g FA), mixed origin FA (39.0 ± 3.2 g/100g FA), preformed FA (37.8 ± 4.8 g/100g FA), and unsaturation index (0.28 ± 0.04 double bonds/FA). Models were evaluated with 10-fold cross-validation and performance was measured with area under the ROC curve (AUC). Data processing, modeling, and evaluation were done using R. The principal components regression yielded AUC of approximately 0.88–0.97, indicating that there is strong separability between samples from lactations 1 and 2 in terms of composition. In particular, de novo FA, preformed FA, and lactose showed the strongest variation, while fat and true protein showed the weakest. Strong differences were found in separability by stage of lactation, with separability in general being strongest in the fresh and early stages, but separability in terms of lactose being strongest in late lactation.

Key Words: lactation, milk composition, milk fatty acid