An approach to reduce computing time in multi-trait single-step evaluations

Multi-trait (MT) single-step (SS) evaluations can be very time-consuming, in particular because of slow convergence. An alternative consists in first running univariate SS evaluations for a limited number of iterations in order to provide reasonable starting values for GEBV or SNP effects as well as corrected phenotypes for fixed effects. This initial step can also include other features to improve the evaluation model such as a correction for heterogenous residual variances. Then a greatly simplified multi-trait SS evaluation can be implemented. Such a strategy was successfully implemented on a Montbeliarde dataset with ~1.6 million animals with performance on a group of 8 correlated type traits.