Goat Milk Nutritional Quality Software-Automatized Individual Curve Model Fitting, Shape Parameters Calculation and Bayesian Flexibility Criteria Comparison

SIMPLE SUMMARY: The high costs of genotyping normally compel researchers to work with reduced sample sizes. Contextually, population observations may no longer compensate for the lack of sufficient data to fit lactation curves, hindering model efficiency, explicative ability, and predictive potential. Individualized lactation curve analyses may save these drawbacks, but may be time-demanding, which may be prevented through computational automatization. An SPSS model syntax was defined and used to evaluate the individual performance of 49 linear and non-linear models to fit the curve described by the milk components of the milk of 159 Murciano-Granadina does selected for genotyping analyses. Protein, fat, dry matter, lactose, and somatic cell counts curves were evaluated and modelled, while peak and persistence were estimated to maximize the ability to understand and anticipate productive responses in Murciano-Granadina goats, which may translate into improved profitability of goat milk as a product. ABSTRACT: SPSS syntax was described to evaluate the individual performance of 49 linear and non-linear models to fit the milk component evolution curve of 159 Murciano-Granadina does selected for genotyping analyses. Peak and persistence for protein, fat, dry matter, lactose, and somatic cell counts were evaluated using 3107 controls (3.91 ± 2.01 average lactations/goat). Best-fit (adjusted R(2)) values (0.548, 0.374, 0.429, and 0.624 for protein, fat, dry matter, and lactose content, respectively) were reached by the five-parameter logarithmic model of Ali and Schaeffer (ALISCH), and for the three-parameter model of parabolic yield-density (PARYLDENS) for somatic cell counts (0.481). Cross-validation was performed using the Minimum Mean-Square Error (MMSE). Model comparison was performed using Residual Sum of Squares (RSS), Mean-Squared Prediction Error (MSPE), adjusted R(2) and its standard deviation (SD), Akaike (AIC), corrected Akaike (AICc), and Bayesian information criteria (BIC). The adjusted R(2) SD across individuals was around 0.2 for all models. Thirty-nine models successfully fitted the individual lactation curve for all components. Parametric and computational complexity promote variability-capturing properties, while model flexibility does not significantly (p > 0.05) improve the predictive and explanatory potential. Conclusively, ALISCH and PARYLDENS can be used to study goat milk composition genetic variability as trustable evaluation models to face future challenges of the goat dairy industry.