Figure 5

Computing time by three parallel computing strategies (a) and prediction correlations by the first strategy (b). Three parallel computing strategies were used in search of optimal SNP panel sizes for predicting genetic merit. The first strategy executed 30 meta-jobs in parallel, each consisting of one round of parallel feature selection (FS) jobs using all SNP markers and one round of parallel post-FS inference and CV for a specific panel (X = 50, 100, 200, …, 3000, respectively); In the second strategy, one round of parallel FS jobs was executed, followed by 30 rounds of parallel post-FS inference and CV jobs conducted sequentially for all panel sizes (P1_A); The third strategy consisted of 30 meta-jobs executed in series, with each meta-job consisting of one round of parallel FS jobs using all SNP markers and one round of parallel post-FS inference and CV for a specific panel (Pn_A).