Genomic prediction in a backcross population using relationship matrices

Abstract. Backcrossing is a widely used crossbreeding strategy for transferring desirable traits from a donor line into a recurrent parent population. Although genomic selection can accelerate genetic improvement in these populations, traditional models such as G-BLUP (genomic best linear unbiased prediction) often assume marker independence and uniform variance contributions – simplifications that can affect the accuracy of genomic estimated breeding values (GEBVs). To address these shortcomings, we developed three models: covariance-adjusted genomic BLUP (CAG-BLUP), which accounts for correlated markers, and two genomic-architecture-specific BLUP variants (GASI-BLUP for independent markers and GASC-BLUP for correlated markers), both assuming unequal variance contributions. Evaluated against the conventional G-BLUP using simulated and empirical mouse datasets, these models demonstrated superior performance in predicting GEBVs and in capturing genetic-architecture nuances. Specifically, GASI-BLUP significantly outperformed G-BLUP in scenarios involving independent quantitative trait loci (QTLs), improving GEBV prediction accuracy by up to 12 % and reducing underestimation of genetic variance by 12 %–34 %. CAG-BLUP showed enhanced performance in dependent QTL scenarios, particularly at lower heritabilities, with an improvement of up to 2 % in GEBV accuracy. These findings highlight the importance of selecting genomic prediction models tailored to the specific genetic architecture of traits to enhance prediction accuracy. By doing so, we pave the way for more effective breeding programs, promising substantial genetic improvements and contributing to the overarching goal of bolstering global food security. Future research will focus on expanding these models to incorporate non-additive genetic effects and on testing their applicability across different species and breeding contexts, aiming to further refine genomic prediction methodologies.