ABSTRACT Accurate knowledge of true digestible amino acid (TDAA) contents of feedstuffs is necessary to accurately formulate poultry diets for profitable production. Several experimental approaches that are highly expensive and time consuming have been used to determine available amino acids. Prediction of the nutritive value of a feed ingredient from its chemical composition via regression methodology has been attempted for many years. The artificial neural network (ANN) model is a powerful method that may describe the relationship between digestible amino acid contents and chemical composition. Therefore, multiple linear regressions (MLR) and ANN models were developed for predicting the TDAA contents of sorghum grain based on chemical composition. A precision-fed assay trial using cecectomized roosters was performed to determine the TDAA contents in 48 sorghum samples from 12 sorghum varieties differing in chemical composition. The input variables for both MLR and ANN models were CP, ash, crude fiber, ether extract, and total phenolswhereas the output variable was each individual TDAA for every sample. The results of this study revealed that it is possible to satisfactorily estimate the TDAA of sorghum grain through its chemical composition. The chemical composition of sorghum grain seems to highly influence the TDAA contents when considering components such as CP, crude fiber, ether extract, ash and total phenols. It is also possible to estimate the TDAA contents through multiple regression equations with reasonable accuracy depending on composition. However, a more satisfactory prediction may be achieved via ANN for all amino acids. The R2 values for the ANN model corresponding to testing and training parameters showed a higher accuracy of prediction than equations established by the MLR method. In addition, the current data confirmed that chemical composition, often considered in total amino acid prediction, could be also a useful predictor of true digestible values of selected amino acids for poultry.

Keywords