Abstract
Genomic selection (GS) has emerged as a promising approach in plant breeding, especially for complex traits like grain yield that are influenced by many small-effect quantitative trait loci (QTLs). Choosing the optimal prediction model is important for GS. Using 4 years of field data from a diverse panel of spring wheat lines, the present study assessed the ability of seven statistical models to predict grain yield (YLD), total spikelet number per spike (tSNS), thousand kernel weight (TKW), plant height (PHT), and heading date (HD). The Reproducing Kernel Hilbert Spaces (RKHS) model was used as a basis for comparing predictive ability improvement of the other six models when major plant adaptation genes controlling flowering time, photoperiod response, plant height, and vernalization were treated as fixed effects. Incorporating fixed effects into the model substantially improved genomic predictive abilities, increasing them by 13.6% for YLD, 19.8% for tSNS, 7.2% for TKW, 22.5% for HD, and 11.8% for PHT. Thus, wheat breeders could apply this approach to GS in cultivar development and improvement. This finding could also be applied for GS in other cereal crops. To our knowledge, this is the first demonstration in spring wheat of integrating adaptive marker sets (FT/Ppd/Rht/Vrn) as fixed effects within an RKHS framework to improve genomic prediction across yield and four yield-related traits.