Development of prediction equations for immunoglobulin A, immunoglobulin G, and immunoglobulin M concentrations in mature milk from Holstein cows using milk infrared spectral data.

Immunoglobulins in ruminant mammary secretions play a central role in active immune protection of the mammary gland against infections. Ig are present in both colostrum and milk from cows, and interest in routinely quantifying the Ig content in milk for herd management and genetic improvement of disease resistance is increasing. Therefore, the objective of this study was to develop a prediction equation for Ig (IgA, IgG, and IgM) concentrations in milk from Holstein cows using milk Fourier-transform infrared (FTIR) spectral data and to evaluate the practical feasibility of the predicted Ig concentration in milk. First, we developed prediction equations for Ig concentrations in milk using 1,633 milk samples comprising both Ig concentrations in milk and milk FTIR spectral data. We then evaluated the predictive accuracy of the developed equations using 3 different factors: derivative preprocessing, spectral wavenumber ranges, and regression models. Our results demonstrated that the prediction equations based on the partial least squares regression and 4 machine learning regression models exhibited the highest predictive accuracy for all traits under the conditions of nonderivative preprocessing and spectral wavenumber range related to milk quality traits. Their predictive accuracies were moderate, with the R ranging from 0.41 to 0.42, 0.50 to 0.52, and 0.38 to 0.39 for IgA, IgG, and IgM, respectively. Second, we evaluated the practical applicability of the predicted Ig concentration by comparing the trends of both the observed and predicted Ig concentrations with respect to several environmental effects. A linear model was applied using the observed and predicted Ig concentrations, and the LSM of the levels for each environmental effect (lactation stage, SCS, parity, and milk yield) was estimated. Our results showed that the estimated environmental effects of the observed and predicted values showed similar trends for all traits. These results indicate that it is possible to estimate environmental effects using the predicted values obtained via the prediction equation with moderate accuracy. Although the predictive accuracy obtained here may be effective for estimating effects at the herd level, further improvement in predictive accuracy is necessary for estimating effects at the cow level.