Combining genomic prediction and multi-trait indices through stochastic simulations: do index type and deployment order affect genetic gain?

Genomic selection (GS) has transformed plant breeding by increasing prediction accuracy and reducing cycle length, but its integration with classical multi-trait selection indices (SI) remains underexplored. In this study, we used stochastic simulations to evaluate seven alternative strategies combining GS with Smith–Hazel (SH), Pesek–Baker (PB), and empirical (EMP) indices in a rice breeding program targeting grain yield (GY), chalkiness rate (CR), and plant height (PH). For index construction, the relative importance assigned to the three target traits was: 70% for increasing grain yield (GY), 15% for decreasing chalky rice (CR), and 15% for maintaining plant height (PH) at a stable level. After a burn-in phase with phenotypic selection, ten recurrent cycles were simulated to compare strategies based on population mean, prediction accuracy, and additive variance. Results showed that the performance of genomic selection (GS) relative to traditional phenotypic indices (TRAD) depends strongly on the target trait and the type of selection index used. Overall, the order, GS or SI first, does not have a significant impact on GS-based and Traditional methods. Also, both the GS and traditional selection methods performed similarly, mainly because the framework and length were the same, even though, in practice, we expect many advantages of the GS-based methods over the traditional ones. Finally, Pesek-Baker provided the more balanced genetic gains among the selection indices, closest to the expectation.