Training genomic selection models across several breeding cycles increases genetic gain in oil palm in silico study

Abstract

Genomic selection (GS) is expected to increase the rate of genetic gain in oil palm. In a GS scheme, breeding cycles with progeny tests (phenotypic selection, PS) used to calibrate the GS predictive model and for selection alternate with GS cycles, making it possible to train the GS model with aggregated data from several cycles. To evaluate this possibility, we simulated four cycles of hybrid breeding for bunch production and compared two methods of calibrating the GS model, one using aggregated data from the two most recent cycles (Tr2Gen), the other using data from the last cycle (Tr1Gen). We also compared a GS scheme with two PS cycles and two GS cycles (2PT-2noPT), and a scheme with PS every other cycle and GS otherwise (PT-noPT). We showed that Tr2Gen had a 10.7% higher genetic gain per cycle than Tr1Gen, mostly due to increased selection accuracy, particularly in across-cycle selection, despite the decreased relationship between training individuals and selection candidates. After four cycles, Tr2Gen had a 5% higher cumulative genetic gain than Tr1Gen, with a lower coefficient of variation. PT-noPT benefited more from the advantages offered by Tr2Gen than 2PT-2noPT. Over four breeding cycles, combining PT-noPT and Tr2Gen largely outperformed conventional reciprocal recurrent selection (RRS), with an increase in annual genetic gain ranging from 37.6 to 57.5%, depending on the number of GS candidates. This study confirms the advantages of GS over RRS and indicated that oil palm breeders should train GS models using all data from past breeding cycles.