Broiler responses to digestible threonine at different ages: a neural networks approach. Issue 4 (6th August 2015)

• Record Type: Journal Article
• Authors: Faridi, A.; Gitoee, A.; Donato, D. C. Z.; France, J.; Sakomura, N. K.
• Is Part Of: Journal of animal physiology and animal nutrition. Volume 100:Issue 4(2016)
• Page Start: 738
• Page End: 747
• Publication Date: 2015-08-06
• Subjects: broiler -- digestible threonine -- neural network; Animal nutrition -- Periodicals; Feeds -- Periodicals; 636.085
• DOI: 10.1111/jpn.12373
• Languages: English
• ISSNs: 0931-2439

Abstract

Summary: Three experiments were conducted with broiler chickens to evaluate the effects of digestible threonine (DThr) and crude protein (CP) on their performance at three different phases of age: 1–14, 15–28 and 29–42 days. The measured traits included the following: average daily gain (ADG), feed intake (FI), feed conversion ratio (FCR), carcass crude protein (CCP), body lipid (BL), feather weight gain (FWG), protein deposited in feather (FCP), carcass plus feather protein (CFCP), carcass Thr deposition (CDThr) and nitrogen excretion (NE). A dilution technique was used to create seven diets (with eight replicates) increasing the DThr content from 1.5 to 10 g/kg of diet for phase 1, 1.3–8.9 g/kg of diet for phase 2, and 1.2–8.2 g/kg of diet for phase 3. Data measured were imported into neural networks (NNs) to: (i) predict the measured traits in response to DThr and CP, (ii) rank the importance of DThr and CP on these traits through sensitivity analysis and (iii) find the optimal levels of DThr and CP that lead to the desired (maximum or minimum) responses. For each trait investigated, 50 different random groups of data were generated using a bootstrapping method. These 50 data groups were then used to develop 50 separate NNs which were subsequently combined to construct the final ensemble NN model. In general, accuracy of the models constructed was acceptable, although models of high (ADG, FCR, CFCP, BL, DThr and NE; 0.64 ≤ R 2 ≤ 0.99) and low (CCP, FWG and FCP; 0.26 ≤ R 2 ≤ 0.79) accuracy were obtained. All models developed showed the greatest sensitivity to DThr. This may be explained by the dilution technique diet preparation used in these experiments. Optimization results showed decreases in optimal values of DThr and CP with increasing age for all traits. The highest level of DThr was suggested for minimum BL, followed by minimum FCR, maximum ADG, maximum CFCP, minimum NE and maximum CCP respectively. Results showed that the optimal values of DThr for minimum FCR in phases 1–3 were 8.5, 7.4 and 6.4 g/kg of diet, while these values for maximum ADG were 8.2, 7.2 and 6.4 g/kg of diet respectively.