Evaluation of prediction equations to estimate gross, digestible, and metabolizable energy content of maize dried distillers grains with solubles (DDGS) for swine based on chemical composition

The objective of this study was to cross-validate prediction equations to estimate the concentration of GE, DE, and ME among sources of corn distillers dried grains with solubles (DDGS) with variable chemical composition in growing pigs. Published concentrations (DM basis) of GE, CP, ether extract (EE), NDF, and total dietary fiber (TDF) along with particle size (PS, μm), bulk density (BD, g/cm3) and in vivo determinations of DE and ME from 45 sources of DDGS samples were obtained from 5 published studies. Prediction equations for GE (5 equations), DE (20 equations), and ME (19 equations) from published studies were used to calculate the concentration of GE, DE, and ME among DDGS sources and compare with experimentally determined in vivo values. Each equation was evaluated using the entire data set, and data sets that excluded data from which the equation was developed (cross-validation). Equations were compared for their overall explanation of variance (R2), precision for reduction in prediction error (PE, kcal/kg DM), and accuracy in deviation of the predicted mean from the overall observed mean (bias, kcal/kg DM). Prediction of GE concentrations among DDGS sources was poor (PE < 200 and biases > 150) despite moderate explanation of overall variance (R2 < 0.6). Therefore, we tested DE and ME equations that included GE as an input using the actual analyzed GE concentration of samples. Under this condition, the most precise (PE = 144) and accurate (bias = 19) DE equation was DE = −2,161 + (1.39 × GE) − (20.7 × NDF) − (49.3 × EE). The most precise (PE = 149) and accurate (bias = −82) ME equation was ME = −261 + (1.05 × GE) − (7.89 × CP) + (2.47 × NDF) − (4.99 × EE). Predicting GE with equation GE = 4,583 + (50.6 × EE) − (0.1 × PS), and using this estimate in the equation of ME = −261 + (1.05 × GE) − (7.89 × CP) + (2.47 × NDF) − (4.99 × EE), resulted in moderate precision (PE = 134) and accuracy (bias = 48). Cross-validation of equations that require PS, BD, or TDF as inputs was not possible because these inputs were only measured in 1 of the 5 published studies used in this evaluation.