Prediction of milk protein concentration from elements of the metabolizable protein system

Elements of the metabolizable protein system in the United Kingdom were examined for their suitability as potential predictors of milk protein concentration. Models were based on data from 163 cows offered five forage mixtures for ad libitum intake plus concentrates at 3, 6, or 9 kg/d of dry matter. The models were then tested on a separate data set of 100 cows offered seven forage mixtures for ad libitum intake plus concentrates at 6 kg/d of dry matter. To minimize problems with collinearity, variables were arranged hierarchically; successive elements were components of variables at higher element levels. Variables from different element levels were not used in the same models. Models were constructed using ridge regression to minimize problems with collinearity. The fit and precision of prediction were generally poor because these models did not take into account animal variables. Models using undergradable dietary protein performed slightly better than did those using digestible undegraded protein. The use of slowly degradable protein and quickly degradable protein rather than rumen-degradable protein generally resulted in improvements in prediction. Models using neutral detergent fiber and quickly fermented carbohydrate were better than those using total carbohydrate. We concluded that there was little to be gained from using the elements of the metabolizable protein system considered here for the prediction of milk protein concentration.