Objective Exploring the difference in prediction accuracy between genome selection (GS) and traditional BLUP methods in the presence of a large number of non-genotype reference populations. Assessing the application effect of GBLUP and ssGBLUP in the joint evaluation to provide recommendations for joint evaluation.

Method Two straits of days to reach 100 kg (DAYS_100) and the average backfat thickness at 100 kg (BFT_100) were analyzed in six Yorkshire populations, and the heritability and genetic correlation of the two traits were estimated. Exploring the prediction accuracy of BLUP, GBLUP and ssGBLUP models in different populations and combined populations.

Result (1) In F population, the heritability of BFT_100 was only 0.071, while in other populations it was between 0.205 and 0.383. The heritability of DAYS_100 in six populations ranged from 0.258 to 0.598. (2) The genetic correlation between the two traits in D population was 0.211, however, in other populations, the genetic correlations were negative, ranging from -0.462 to -0.200. (3) For DAYS_100 trait, the GBLUP model showed the best prediction accuracy in B, C, E, and F populations. For BFT_100 trait, the ssGBLUP model had the best prediction accuracy in populations A, B, and C, while the GBLUP model performed better in populations D and E. (4) The connectedness rating (CR) between F and A population was 3.096 %. In F population, single-step genomic selection using combined reference populations can improve the prediction accuracy of BFT_100 trait.

Conclusion When the number of genotyped individuals in the population exceeds 500 and the proportion is above 7%, the prediction accuracy of GBLUP or ssGBLUP model will be higher than that of BLUP model. ssGBLUP model can be used to improve the prediction accuracy of low heritability traits in joint population that the CR reaches 3% between population.