Whole-genome sequencing captures a wide spectrum of genomic variations beyond single nucleotide polymorphisms (SNPs). Harnessing these diverse variations could be beneficial for genomic prediction. This study investigates the influence of SNPs and insertion-deletions (InDels) on genomic prediction within the Russian sturgeon (Acipenser gueldenstaedtii) population, focusing on key traits such as caviar yield, caviar color, and body weight. We generated whole-genome sequencing data comprising 10,409,793 high-quality SNPs and 4,938,138 InDels from 673 sequenced individuals (~13.68×). Using linkage disequilibrium pruning, we reduced whole-genome sequencing data to approximately 50K markers, facilitating an evaluation of various prediction methods. Our results reveal that prediction accuracy improves with increased marker density, peaking around 50K markers, and that the inclusion of InDels from whole-genome sequencing further enhances prediction accuracy for specific traits. The multi-source BLUP (MSBLUP) method, which integrates both SNPs and InDels, achieves an average prediction accuracy 1.5% higher than GBLUP. Additionally, Bayesian methods demonstrated comparable prediction accuracy to GBLUP. Kernel ridge regression achieved the highest prediction accuracy, with an average improvement of 2.2% over GBLUP across all traits assessed. Overall, these results highlight the benefits of utilizing a diverse set of genetic markers and advanced prediction methods to enhance genomic prediction accuracy in aquaculture breeding programs.

Genomic prediction,linkage disequilibrium pruning,Bayesian methods,machine learning methods,sturgeon