Optimisation of genomic selection for harvest traits of Malabar red snapper (Lutjanus malabaricus)

Selective breeding in aquaculture has traditionally relied on pedigree-based selection to improve economically important traits. However, these methods have limitations, including lower accuracy, pedigree errors, and an inability to fully capture and exploit within-family genetic variation. Genomic selection (GS) offers a more precise alternative by leveraging genome-wide single nucleotide polymorphisms (SNPs) to calculate genomic estimated breeding values (GEBVs). This study aimed to optimize GS for Malabar red snapper ( Lutjanus malabaricus ), a commercially valuable species in Southeast Asia. We evaluated different SNP densities, training population sizes, and three GS models: GBLUP (genomic best linear unbiased prediction), BayesR (a hierarchical Bayesian model), and KAML (kinship-adjusted multiple-loci linear mixed model). A total of 2547 snappers from three rearing sites were phenotyped and genotyped at harvest (~1.5 years of age) using a custom SNP array. Five traits—body weight (BW), total length (TL), body depth (BD), Fulton's condition factor (K), and body shape index (BSI)—were analysed using six SNP densities (500, 1000, 5000, 10,000, 30,000, and 56,378 SNPs) selected randomly, or ranked based on genome-wide association study (GWAS) p -values on body weight. We also tested six training population sizes (80, 400, 800, 1200, 1600, and 2038 individuals). Prediction accuracy was assessed via five-fold cross-validation with 10 replicates. Results showed that prediction accuracy improved with larger training population sizes and higher SNP densities, plateauing at 1200 to 1600 training population size and 5000 SNPs for growth traits. GWAS-ranked SNP subsets significantly outperformed corresponding random subsets in predicting GEBVs for body weight when the training and test populations were from the same dataset. GBLUP and KAML achieved similar prediction accuracies for BW and TL (0.45–0.50) and outperformed BayesR (0.40–0.45). In contrast, BayesR performed better for the other traits, yielding the highest accuracies for BD (0.62), K (0.78) and BSI (0.71) using the full dataset. GEBVs from GBLUP and KAML were highly correlated across all traits (≥0.95), whereas BayesR showed strong correlations with GBLUP and KAML for BW and BD (>0.93), moderate correlations for TL and BSI (0.54–0.74), and no significant correlation for K. These findings suggest that using 5000 SNPs and a training population of 1200–1600 individuals provide a cost-effective balance between accuracy and efficiency for growth traits when using the GBLUP model. This study offers key insights for genomic breeding programs in Malabar red snapper, contributing to the sustainability and productivity of aquaculture in Southeast Asia. • Optimized genomic selection for red snapper using SNP panels and GS models. • Identified cost-effective SNP density and training population size for GEBV accuracy. • GWAS-ranked SNPs improved prediction but effects varied by trait. • Compared GBLUP, BayesR, and KAML to guide trait-specific selection. • Enables efficient genomic breeding for tropical marine aquaculture.