Genomics and Applied Biology 2024, Vol.15, No.6, 307-319 http://bioscipublisher.com/index.php/gab 312 additive genetic architecture and can model non-additive genetic effects, which are common in complex traits. Nonparametric models are especially useful for capturing complex interactions between markers and environmental variables, although they require large datasets and extensive computational power (Zhao et al., 2020). Altogether, the choice of model depends on the trait architecture, computational resources, and specific breeding goals. 5.2 Comparative effectiveness of different models in yam genomic selection Different genomic prediction models demonstrate varying levels of effectiveness depending on the trait being studied, the genetic architecture, and the available data. In yam breeding, where traits like yield, disease resistance, and tuber quality are highly complex, the choice of an effective prediction model is critical. Studies comparing the performance of models, such as GBLUP and Bayesian methods, suggest that Bayesian models often outperform GBLUP for traits controlled by a few major loci due to their flexibility in handling marker-specific effects. For example, in traits where non-additive genetic effects are substantial, Bayesian models like Bayes B and Bayes Cπ have shown better predictive accuracy because they allow some markers to have larger effects than others. From the study by Asfaw et al. (2020), it can be seen that the genetic variance composition differs among traits. For example, the dominance component is higher for tuber yield and tuber number, while the additive component is lower for fresh tuber yield. This information helps in understanding the genetic control patterns of different traits (Asfaw et al., 2020) (Figure 1). However, GBLUP remains widely used and is considered robust for a broad range of traits, particularly polygenic ones where many genes with small effects collectively contribute to the trait. GBLUP’s simplicity and computational efficiency make it a practical choice, especially for breeders working with limited resources. Machine learning models, such as random forests, support vector machines, and deep learning approaches, have shown promising results in other crops for predicting traits with complex architectures. In recent research, deep learning models like convolutional neural networks (CNN) and multilayer perceptrons (MLP) have outperformed traditional models for some complex traits in crops such as wheat (Sandhu et al., 2021). These models could offer similar benefits in yam, where complex traits like yield and disease resistance involve intricate gene-environment interactions. Figure 1 Genetic variance distribution of six traits in white Guinea yam (Adopted from Asfaw et al., 2020)
RkJQdWJsaXNoZXIy MjQ4ODYzMg==