International Journal of Aquaculture, 2025, Vol.15, No.2, 76-87 http://www.aquapublisher.com/index.php/ija 81 computing power of cloud servers. Another important aspect is the application of big data and artificial intelligence technologies in breeding decision-making. Analyzing multi-generation breeding data through machine learning algorithms can optimize the optional solutions, control the growth of in-kin coefficients, and predict genetic progress under different breeding schemes. Big data analysis can be used to identify key environmental factors that affect the growth and survival of abalone, thereby guiding the precise delivery of good varieties in different breeding areas. 5 Research Progress of Genome Selection (GS) Technology in Abalone 5.1 Analysis of the applicability of GS principle in abalone Genome selection uses molecular markers to cover the entire genome to predict individual breeding values, which have achieved remarkable results in domestic livestock, poultry and some aquatic animals. For abalone with longer generation intervals, the application of GS has a clear potential advantage. Wrinkle disc abalone is generally sexually mature in 2 to 3 years. If traditional phenotype breeding is adopted, the selection cycle of one generation often takes more than 3 years. GS allows early selection based on genotype at the larval stage, thereby shortening generational intervals and improving genetic progress. Secondly, the abalone has strong fertility (single females can lay millions of eggs), and a large enough reference group can be established to support GS model training. Taking heat tolerance as an example, one study assessed heat tolerance phenotypes in over a thousand Haliotis diversicolor squamata (Wrinkled Disk Abalone) individuals and conducted a genome-wide scan, and estimated that the breeding value was as accurate as possible. This accuracy has been significantly higher than the traditional genealogical BLUP method, greatly increasing the choice credibility. Some target traits of abalone (such as disease resistance) cannot be directly measured on candidate abalone, and the species-using individual must be selected through family measurement of siblings. GS can use data on survival or disease-challenged individuals, train the model and then predict the disease-resistant breeding value of healthy abalone, thereby achieving the selection of untested individuals, which is particularly important for improving disease-resistant breeding efficiency. Of course, some special factors need to be considered in the application of GS: the cost of abalone larvae cultivation and family construction is high, and the size of the training group will affect the accuracy of the model; the environment of different abalone breeding varies greatly, and the GS model may have an extrapolation risk of environmental adaptability. 5.2 Construction and evaluation methods of prediction model The core of genomic selection is to build reliable genomic breeding value (GEBV) prediction models. Currently in the abalone GS study, the main statistical methods used include GBLUP (Genome Optimal Linear Unbiased Prediction) and various Bayesian models. GBLUP uses all markers to construct a genomic affinity matrix to directly predict GEBV, and is computationally efficient and suitable for traits that are mainly additive genetic. Bayesian model assumes that different markers have different effect distributions, which can give greater weight to a few large-effect QTLs, and may perform better when traits are affected by major genes. Data simulation studies on abalone show that Bayesian models have better prediction accuracy than GBLUP when there are a few large effect sites; however, in real data applications, Bayesian methods often perform comparable or slightly advantageous to GBLUP. In terms of model evaluation, cross-validation or independent validation set methods are usually used to measure prediction accuracy, that is, to calculate the correlation coefficient of GEBV to the actual phenotype (or traditional breeding values). The reference population is randomly divided into training sets and verification sets, and multiple cycles of verification can be obtained to know the average correlation coefficient of the model on unknown individuals. In the example of abalone heat resistance traits, the average accuracy of the BayesB model with 5-fold cross-validation reached 0.55±0.05, indicating that the model has high robustness (Liu et al., 2022). Furthermore, direct evaluation of the selection response can be used, i.e. the phenotypic advantage of species abalone selected according to GEBV compared to unselected populations.
RkJQdWJsaXNoZXIy MjQ4ODYzNA==