Pea genomic selection for Italian environments

BMC Genomics

Table 8 Correlation (r) of phenotypic data or genomic selection (GS)-modelled data based on two (test) environments with phenotypic data in another (validation) environment, square root of broad-sense heritability (H) when adopting two test environments, accuracy (r_Ac) of GS modelling trained in two environments for prediction in one validation environment, and GS vs phenotypic selection (PS) efficiency ratio based on predicted genetic gains per unit time for similar evaluation costs assuming two environments for PS and for generation of phenotyping data for intra-population (GS_A) and inter-population (GS_B) GS scenarios, for four pea traits. Data averaged across three environment combinations and two (lodging susceptibility) or three (other traits) RIL populations

^a GY, grain yield; OF, onset of flowering; LS, lodging susceptibility; SW, individual seed weight
^b Using Bayesian Lasso modelling trained on all genotype data
^c Assuming experiments with three replicates (as the current phenotyping experiments)
^d Using Bayesian Lasso modelling for prediction of independent lines, using 50 repetitions of 10-fold stratified cross-validation per individual analysis
^e As ratio (i_G r_Ac / t_G) / (i_P H / t_P), where i_G and i_P are standardized selection differentials for GS and PS, respectively, setting i_G = 1.46 i_P to approach same evaluation costs; and t_G and t_P are cycle duration for GS and PS, setting t_G = 0.5 and t_P = 1 (two test sites in the same test year) or t_P = 2 (two test years in the same site)
^f Using Bayesian Lasso model training on data of one RIL population for prediction within each of two other populations

ISSN: 1471-2164