Accurate prediction of feed intake (FI) and feed conversion ratio (FCR) is crucial for industry nutritionists. This study aimed to develop models to predict FI and FCR of broilers based on dietary Arkansas Net Energy (ArkNE), metabolizable (AME, AMEn), and classic net (CNE) energy values. Four experiments (E1 to E4) fed broilers a total of 60 pelleted diets within 1-56 d (four experiments, five phases each, three treatments per phase) following Cobb 500 guidelines. They included replacements of soybean meal with cellulose (E1) or corn (E2) and inclusion levels of oil (E3) or total digestible amino acids (E4). A total of 9600 birds were used (20 birds/pen, eight pens –the experimental unit– per treatment, three treatments per feeding phase, five phases assessed, four experiments). Body composition, heat production, and performance were assessed, and data analyzed in JMP (experiment = random factor) with a random data-splitting validation and refitting approach. Models were assessed (normality, homoscedasticity) and compared based on their precision (adjR2), error (RMSE), certainty (energy content P value), and consistency among experiments. The FI (g/bird/d) and FCR were natural log transformed (Ln) to reduce heteroscedasticity. ArkNE was the only system showing an inverse and consistent correlation with FI in all experiments (the higher the ArkNE, the lower the FI). The model to predict FI based on ArkNE (kcal/kg) as a predictor (P<0.001) was: Ln FI = – 2.77 – 0.000097 ArkNE + 0.038 age – 0.00159 (age – 31.1548)2 (adjR2=0.94). The FI was insensitive to the diet AME (P=0.377) or AMEn (P=0.427). The relationship between diet CNE and FI was not only inconsistent among studies but also positive in three of them and overall (the higher the CNE, the higher the FI). Models predicting FCR showed an equivalent pattern. ArkNE was the only system showing an inverse and consistent correlation with FCR among experiments. The model to predict FCR based on ArkNE as a predictor (P<0.001) was: Ln FCR = 0.6646 – 0.000225 ArkNE + 0.0112 age (adjR2=0.81). The relationship between FCR and diet AME or AMEn was inconsistent (positive in two studies; negative in two). The relationship between diet CNE and FCR was not only inconsistent among studies but also positive in three of them and overall. Indeed, a single model to predict Ln FCR, including ArkNE, AME (or AMEn), CNE, and age as predictors (adjR2=0.81; no multicollinearity among systems, VIF<4), showed a highly significant influence of the ArkNE (P<0.001) on the FCR but no influence of AME (P=0.768), AMEn (P=0.678), or CNE (P=0.662) on FCR. All models were validated.
In conclusion, ArkNE was consistently a better predictor of FI and FCR than alternative energy systems.