With regard to parameter estimates, Schulthess et al. 2013).Īs mentioned above, multiple-trait analysis improves parameter estimates and prediction accuracy. Furthermore, implementing the second approach is also challenging because early breeding programs start with at least 1000 lines that are evaluated in multiple environments, which complicates the analysis, as including genotype × environment (G × E) increases the dimensionality of the data considerably (Chiquet et al. However, implementing the second approach is sometimes challenging-for example, when there are a large number of traits, and under these circumstances, breeders opt for the first approach.
Implementing first approach is valid when the correlation between traits is low or close to zero, but it is less desirable when the correlation between traits is moderate to strong. To analyze this type of experiment, breeders implement one of the following two approaches: (i) they perform a univariate analysis (one trait at a time), therefore ignoring the correlation between traits that does not allow to improve either parameter estimates or prediction accuracy, or (ii) they perform a multiple-trait analysis, which may not only take into account the correlation between traits but may also significantly increase the computing intensity. ( 2016) reported that a breeding program in Brazil measured 41 pig traits obtained by crossing 345 F2 pig populations of Brazilian Piau × commercial pigs.
Moreover, the proposed multiple-trait method is atractive because it can be implemented using current single-trait genomic prediction software, which yields a more efficient algorithm in terms of computation.īreeders often want to improve more than one trait simultaneously in their breeding programs and thus conduct various experiments. We found that the proposed method based on SVD produced similar results, in terms of parameter estimation and prediction accuracy, to those obtained with the BMTME model. The results of the proposed method are compared, in terms of parameter estimation and prediction accuracy, with the results of the Bayesian multiple-trait and multiple-environment model (BMTME) previously described in the literature. In stages three and four, we collect and transform the traits back to their original state and obtain the parameter estimates and the predictions on these scale variables prior to transformation. In the first stage, we perform singular value decomposition (SVD) on the resulting matrix of trait responses in the second stage, we perform multiple trait analysis on transformed responses. Consequently, we propose a four-stage analysis for multiple-trait data in this paper.
However, frequently there are several traits under study at one time, and the implementation of current genomic multiple-trait and multiple-environment models is challenging. Today, breeders perform genomic-assisted breeding to improve more than one trait.