Uma abordagem computacional para estudo comparativo das estimativas de componentes de variância e parâmetros genéticos sob enfoque bayesiano em relação à abordagem clássica por máxima verossimilhança

Abstract

The genetic evaluation of animals is based on analytical methods that combine multidisciplinary knowledge, especially genetics, statistics and computer science. Genetic parameters can be obtained by these methods via several software applications; many of which interact with the user only through text files and command line. The first aim of this study was to provide the researcher with a user-friendly operating platform to integrate various data analysis applications in a single graphical interface, assisting and easing the use and comparison of different methodologies; the second was to compare the employment of two analytical methods for the genetic evaluation of quails, using the facilities of this platform. The developed computational tool was set up and is available in the form of two operating assistants. Different functions are offered including: a graphical user interface, friendly, generic and configurable, to integrate different scientific applications of data analysis, giving them increased usability; resources for planning, organizing, executing, registering and recovery of analyses; facilities for data manipulation (collects and preparation) and visualization (input data and reports). This tool was used integrated with the WOMBAT and INTERGEN applications for the genetic evaluation of 5726 quails, through the classical and bayesian methods. Data were collected from 10 generations of a breeding program and refer to body weight measured in seven ages - one, seven, 14, 21, 28, 35 and 42 days - and to pedigree. Univariate analysis were applied to several data sets relative to each age, grouping 7, 8, 9 or 10 successive generations. The estimates indicate that: the heritability (narrow sense) is higher at 35 days (h2 = 0.46 for the full group); there is a slight reduction of heritability for all ages, as the generations 8, 9 and 10 are included; the variance components have a high degree of persistence across generations 7 to 10. It is evidenced by comparisons between the corresponding genetic parameters that both methodologies are equivalent. The classical approach using the REML methodology is satisfactory for obtaining good estimates of the parameters.