Genomic predictions improve the performance of clonal cultivars in oil palm. [PE0834]

Prediction of clonal genetic value is among the difficulties of the genetic improvement of oil palm (Elaeis guineensis Jacq.) yield. Presently, clonal selection requires two stages of phenotypic selection (PS): preselection on the phenotypic values of one or two yield components having high heritability, and final selection on performances in clonal trials. The current study evaluated the efficiency of genomic selection (GS) for clonal selection on eight traits. The GS models were trained on 295 and 279 Deli × La Mé crosses for bunch production and quality components, respectively, and were validated on 42 Deli × La Mé ortets of known clonal value. Genotyping by sequencing led to a dense genome coverage with 15,054 single nucleotide polymorphisms (SNP). We assessed the effects of SNP dataset (SNP density and quality) and of two GS modelling approaches on prediction accuracy. The results showed prediction accuracies that ranged between 0.70 and -0.03 according to trait, SNP dataset and model. Modeling disregarding the parental origin of alleles was preferable given the simplicity of implementation and the robustness over traits and SNP datasets, although including parental origin effects could slightly increase prediction accuracies for the traits used to define the two oil palm heterotic groups (bunch number and average bunch weight). The greatest GS prediction accuracies were beyond those of PS for most of the traits. Prediction accuracies from 0.70 to 0.45 for all traits can be achieved combining GS and PS. The best GS prediction accuracies are achieved with at least 7,000 SNPs. This will enable preselecting ortet candidates on all traits before clonal trials, thus increasing the selection intensity and the genetic progress