The integration of broiler chicken threonine responses data into neural network models

In making general recommendations for amino acids, researchers might survey various published data on the responses of poultry to amino acids. In this way, the use of appropriate mathematical and statistical approaches may help researchers draw appropriate conclusions. The purpose of this study was to develop artificial neural network (ANN)-based models to analyze data on the responses of broiler chickens [ADG and feed conversion ratio (FCR)] to protein and threonine from 21 to 42 d of age. A data set containing 92 dose-response treatments was extracted from the literature, compiled, and entered into the training and testing sets of the ANN models. The constructed models were subjected to a process of sensitivity analysis to evaluate the relative importance of the effects of dietary protein and threonine on ADG and FCR. Optimal values for the input variables (protein and threonine requirements) to maximize ADG and minimize FCR in birds were obtained by using the ANN models with an optimization algorithm. Based on the calculated goodness of fit criteria, it appeared that the platform of ANN-based models with the sensitivity analysis and optimization algorithms was an efficient tool for integrating published data on the responses of broiler chickens to threonine. The analyses of ANN models for ADG and FCR based on the compiled data set suggested that the dietary protein concentration was more important than the threonine concentration. The optimization algorithm revealed that diets containing 18.69% protein and 0.73% threonine could lead to optimal ADG, whereas the optimal FCR could be achieved with diets containing 18.71% protein and 0.75% threonine.