Predicting for disease resistance in aquaculture species using machine learning models

Predicting disease resistance is one of the most prominent applications aquaculture selective breeding. Reductions in genotyping costs have allowed implementation genomic selection an abundance species and their related diseases showing promising results. Machine learning (ML) models can be value for prediction purposes, as suggested by several studies both plants livestock. The current study aimed to test efficiency various ML predicting using simulated real datasets. More specifically, like decision trees (DT), support vector machines (SVM), random forests (RF), adaptive boosting (Adaboost) extreme gradient (XGB) were benchmarked against best linear unbiased threshold traits backend Markov chain Monte Carlo (GBLUP-MCMC) terms required computational time. Moreover, model ranking was tested datasets where ratio between two observed phenotypes (resistant vs non-resistant) unbalanced. Across all datasets, XGB ranked first with a slight advantage over GBLUP-MCMC, ranging 1–4 %. SVM RF delivered predictions tight proximity ones from GBLUP-MCMC. In addition, 3–4 % lower compared GBLUP-MCMC obtained Adaboost. On other hand, DT consistently low (?40 GBLUP-MCMC). All had significantly reduced requirements than case XGB, more 20-fold opposed under settings study. competitive highly efficient time (?3 min). Overall, results suggest that valuable tools breeding resistance.