Wheat is a common raw material used to provide most of the energy and a great portion of amino acids in the poultry diet. The routine investigation of metabolizable energy (ME) and digestible amino acid content determination are too costly and time consuming for wheat grains. Therefore it would be helpful if the and digestible amino acid content of wheat grain samples could be predicted from their chemical compositions. Three studies were conducted to evaluate the probability of AMEn, AME and apparent ileal digestible amino acid (AIDAA) prediction in wheat samples based on chemical compositions. Multiple linear regression (MLR), partial least square (PLS) and Artificial neural network (ANN) methods were developed to estimate the AME values of wheat grain samples based on total and soluble non-starch polysaccharides (study 1) and the AMEn based on dry matter (DM), crude protein (CP) and Ash (study 2). Furthermore MLR and ANN models were used to estimate the AIDAA via CP content of wheat samples (study 3). The fitness of the models in each study was tested using R2 values, RMS error, MAD, MAPE, and bias parameters. The results of studies 1 and 2 showed that AME can be predicted from the chemical compositions. The prediction of AME of wheat through ANN-based model showed higher accuracy and lower error parameters as compared to MLR and PLS models in both studies (1 and 2). The results of 3rd study indicated that CP can be used as a single model input to predict AIDAA in wheat samples. Furthermore ANN model may be used to improve models performance to estimate AIDAA as affected by CP content. The results of this study demonstrated that the ANN model may be used to accurately estimate the ME and AIDAA values of wheat grain from its corresponding chemical compositions. As a result, this method provides an opportunity to reduce the risk of an unbalanced level of energy and amino acid, in feed formulation for poultry.

Keywords