Genomic prediction of survival traits for scuticociliatosis resistance in a vaccinated olive flounder cohort: Comprehensive evaluation and optimization of statistical and machine learning models

Genomic prediction utilizes genome-wide markers to facilitate genomic selection in aquaculture, enabling more precise and accelerated genetic gains compared to traditional pedigree-based methods. Training population size and marker density influence genomic prediction accuracy, and both contribute to the overall cost of selection programs. Scuticociliatosis, caused by the parasite Miamiensis avidus , poses a major threat to olive flounder aquaculture and can be managed through selective breeding. We evaluated the effects of marker density and training population size on prediction accuracy using 10 genomic prediction models, including pedigree-based best linear unbiased prediction (BLUP PBLUP), genomic BLUP (GBLUP), Bayesian methods (BayesA, BayesB, BayesC, Bayesian Lasso, Bayesian Ridge Regression), regularized regression (Ridge Regression and Elastic Net EN), and random forest (RF). The analysis was conducted using 474 genotyped individuals comprising 60 paternal half-sib families. Model performance was evaluated using 5-fold cross-validation repeated 10 times. GBLUP and Bayesian methods consistently outperformed PBLUP, EN, and RF across all survival traits. GBLUP achieved a significantly higher prediction accuracy (0.64–0.72), which is more than twice that of PBLUP (0.06–0.31). The highest predictive ability (0.558 ± 0.006) was achieved with the top 1000 GWAS-ranked markers for the GBLUP model, highlighting the importance of informed marker selection. Contrastingly, random marker selection exhibited no clear gains in predictive performance. Predictive ability was improved by increasing training population size, with no significant differences between 3- and 5-fold cross-validation at larger sample sizes ( p < 0.001). These findings underscore the importance of selecting appropriate genomic prediction models tailored to the genetic architecture of survival traits, and the benefit of GWAS-informed marker selection to maximize genomic prediction ability in olive flounder aquaculture breeding programs. • Genomic prediction of scuticociliatosis resistance was performed using a 70 K SNP array. • GBLUP and Bayesian models outperformed PBLUP, Elastic Net and Random Forest models. • GWAS-informed marker selection improved predictive ability over the random approach. • Larger training populations enhanced predictive ability until performance plateaued.