Empirical Prediction Accuracy of Genomic Selection Between Experimental Designs and Generations in Oil Palm

There is a large potential for the genetic improvement of yield in oil palm, as it has only been submitted to a few generations of modern breeding. Selection candidates are traditionally evaluated in progeny tests, as some yield components have a low heritability. The progeny-tests allow selecting with high accuracy, but constrain the rate of genetic gain by increasing the generation interval and limiting the number of evaluated individuals. Genomic selection (GS) is an appealing alternative that could allow selecting without progeny tests.

We studied the prediction accuracy of GS, ie the correlation between the genomic estimated breeding values and the breeding values obtained from the conventional selection method, in the two heterotic groups used to produce commercial hybrids. We used two independent experimental designs located in Sumatra (Indonesia), comprising over 700 hybrid crosses. The GS model was calibrated with the first experimental design (training set) and applied to predict the breeding values of the individuals progeny tested in the second design (validation set). The genomic breeding values were obtained by GBLUP, using the phenotypes of the training hybrids and the genotypes of all the progeny tested individuals. The genotypes consisted in 3000+ SNPs from genotyping-by-sequencing (GBS).

We found that it was possible to achieve satisfactory GS accuracy provided the set of SNPs was carefully defined (density and filtering criteria). Finally, GS appeared as a highly valuable approach to speed up the genetic progress in oil palm yield, and GBS as a suitable technology to generate the genotypes required.