Bioeficácia do ácido 2-hidróxi-4-metiltiobutanoico e exigências de metionina + cistina sobre o empenamentoem frangos de corte, analisadas por imagens digitais assistidas por IA

Abstract

The objectives were to estimate the relative bioefficacy of 2-hydroxy-4-methylthiobutanoic acid (HMTBa) compared to DL-methionine (DLM), to determine digestible methionine + cystine (Met+Cys) requirements, and to validate an artificial intelligence-assisted digital morphometry method for feathering quantification in male Cobb 500 broilers during the starter (1–18 days), grower (18–32 days), and finisher (32–42 days) phases. For the bioefficacy studies, three independent experiments were conducted at the Poultry Research Facility of the Federal University of Goiás (Goiânia, GO, Brazil), totaling 5,664 birds, in a completely randomized design with 11 treatments: a methioninedeficient basal diet (6 replicates) and ten diets supplemented with increasing levels of DLM or HMTBa (0.04, 0.08, 0.15, 0.25, and 0.35%), on a weight-to-weight basis (9 replicates each), with 23, 20, and 16 birds per experimental unit in the starter, grower, and finisher phases, respectively. Relative bioefficacy was estimated using five simultaneous regression models — exponential, linear (slope-ratio), quadratic, Linear Response Plateau (LRP), and Michaelis- Menten — fitted to individual body weight and weight gain data in the R environment. In the starter phase, nonlinear models estimated HMTBa bioefficacy between 73 and 79% for weight gain, with significant differences detected by the LRP (73.09%; p = 0.025) and the exponential model for body weight (75.92%; p = 0.048). In the grower phase, bioefficacy ranged from 79 to 88% with no significant difference in any model (p > 0.05). In the finisher phase, bioefficacy ranged from 92 to 100%, with the Michaelis-Menten model estimating 99.72%, indicating virtually complete functional equivalence. The linear model systematically overestimated bioefficacy (92–113%) and violated the residual normality assumption across all three phases. The LRP model was recommended as the reference due to its greater sensitivity in detecting differences between sources across all phases. Phase-specific correction factors were proposed: 1.35× in the starter, 1.17× in the grower, and 1.0× in the finisher phase. For the determination of nutritional requirements, three experiments were conducted with 1,173, 1,020, and 816 birds in the starter, grower, and finisher phases, respectively, in a completely randomized design with six treatments corresponding to increasing levels of digestible Met+Cys (0.578 to 0.920% in the starter; 0.523 to 0.865% in the grower; 0.484 to 0.826% in the finisher), obtained by the dilution method between a basal diet without DLM and a concentrated diet containing 0.35% DLM. Requirements were estimated using the Quadratic Response Plateau (QRP) regression model in the R environment (EASYREG package). In the starter phase, the QRP model (R² = 0.997 for body weight and weight gain; R² = 0.954 for feed conversion ratio) estimated requirements of 0.860%, 0.862%, and 0.854% for body weight, weight gain, and feed conversion ratio, corresponding to Met+Cys:Lysine ratios of 69%, 69%, and 68%. In the grower phase (R² = 0.985; 0.995; 0.988), requirements were 0.727%, 0.760%, and 0.781%, with ratios of 66%, 69%, and 71%. In the finisher phase (R² = 0.926; 0.876; 0.936), requirements were 0.705%, 0.701%, and 0.736%, with ratios of 72%, 72%, and 75%. For the digital morphometry validation, two experiments were conducted in the starter (1,173 chicks) and grower (1,020 birds) phases, using the same six DLM treatments. At the end of each phase, two birds per experimental unit were photographed using a Canon EOS Rebel T7 DSLR camera under a standardized protocol (ISO 200, f/8.0, 1/60 s, 50 cm distance, diffuse lighting). Images were processed through an automated Python pipeline assisted by the ChatGPT-4 platform (OpenAI), comprising GrabCut segmentation, dorsal ROI definition, grayscale conversion, binarization with a fixed threshold (T = 210), and quantification of the percentage of pixels classified as feather. At 18 days, the QRP model (R² = 0.934; p = 0.0168; CV = 16.00%) estimated a critical point at 0.825% digestible Met+Cys with a plateau of 5.14% coverage. At 32 days, the QRP model (R² = 0.947; p = 0.0122; CV = 4.61%) estimated a critical point at 0.843% with a plateau of 34.82%. The method demonstrated sensitivity to detect significant differences in feathering among nutritional treatments in both phases. In conclusion, HMTBa bioefficacy relative to DLM shows a consistent ontogenetic progression (~76% in the starter, ~83% in the grower, and ~97% in the finisher phase), consistent with the maturation of hepatic enzymatic conversion systems; digestible Met+Cys requirements progressively decrease with age, with Met+Cys:Lysine ratios of 68%, 71%, and 75% recommended for the starter, grower, and finisher phases, respectively, based on feed conversion ratio; and artificial intelligence-assisted digital morphometry constitutes a viable, reproducible, and complementary tool for objective feathering quantification in broiler chickens.