Prediction of ruminal volatile fatty acid proportions of lactating dairy cows based on milk odd- and branched-chain fatty acid profiles: new models, better predictions.

The volatile fatty acids (VFA) produced in the rumen and the proportions in which they are produced are important determinants of a ruminant's metabolism, but their monitoring requires rumen-fistulated animals, which is not feasible under practical conditions or in experimental setups at herd level. An alternative approach was suggested earlier, consisting of predicting the VFA proportions from measured odd- and branched-chain fatty acid concentrations in the milk with a linear model. Here, we have improved this strategy through the development and application of 2 new model structures: the quadratic model, containing quadratic terms and interactions, and the rational model, consisting of a ratio of linear expressions. Both were found to improve prediction accuracy significantly compared with the linear model. Although the quadratic model achieved the best prediction accuracy, the rational model has the interesting property that it takes the dependence of the 3 predicted VFA into account and guarantees that the 3 proportions add up to 1. Adding a study effect to correct for a possible study bias in the multi-study data improved prediction substantially for all 3 methods. Our results demonstrate the potential of using milk odd- and branched-chain fatty acid concentrations to predict rumen VFA proportions.