Molecular Plant Breeding 2024, Vol.15, No.5, 247-258 http://genbreedpublisher.com/index.php/mpb 252 Incorporating QTL information into GS models can significantly enhance prediction accuracy by ensuring that the markers are closely linked to the traits of interest. This approach captures both major and minor genetic effects, resulting in more reliable breeding value predictions (Goddard and Hayes, 2007; Desta and Ortiz, 2014; Meuwissen et al., 2021). For example, the StepLMM model, which integrates QTL mapping with GS, has demonstrated higher prediction accuracy compared to traditional GBLUP models by effectively utilizing significant SNPs identified through QTL mapping (Li et al., 2017). Additionally, including whole-genome sequence data in genomic prediction models further improves accuracy by capturing fine-scale genetic variations (Meuwissen et al., 2021). 5.2 Case studies and examples Several studies have successfully integrated QTL mapping and GS across various species, demonstrating the potential of this approach in breeding programs. For instance, in dairy cattle, the combination of whole-genome sequence data from different breeds and Bayesian genomic prediction has improved prediction accuracies and enabled the fine-mapping of QTLs associated with milk production traits (Meuwissen et al., 2021). Similarly, in plant breeding, GS models incorporating QTL information have shown higher prediction accuracies for traits such as yield and disease resistance (Desta and Ortiz, 2014). In E. ulmoides, the integration of QTL mapping and GS holds significant promise for improving growth traits and other economically important characteristics. Recent studies have constructed high-density genetic maps and identified numerous QTLs associated with growth traits in E. ulmoides (Li and Sillanpää, 2012; Jin et al., 2020; Liu et al., 2022). Incorporating these QTL data into GS models can improve the accuracy of breeding value predictions and accelerate the development of superior E. ulmoides varieties. For example, the identification of 44 QTLs related to growth traits provides a solid foundation for implementing GS in E. ulmoides breeding programs (Liu et al., 2022). 5.3 Future directions The future of integrating QTL mapping and GS lies in the adoption of emerging technologies and methodologies. Advances in high-throughput sequencing and genotyping technologies will enable the generation of more comprehensive genomic data, facilitating the identification of additional QTLs and improving the resolution of genetic maps (Meuwissen et al., 2021; Liu et al., 2022). Moreover, the development of more sophisticated statistical models, including those incorporating machine learning algorithms, will enhance the predictive power of GS models (Li and Sillanpää, 2012; Li et al., 2017). Integrating multi-omics data, such as transcriptomics, proteomics, and metabolomics, with QTL mapping and GS represents a promising direction for future research. By combining different layers of biological information, researchers can achieve a more holistic understanding of the genetic and molecular mechanisms underlying complex traits (Goddard and Hayes, 2007; Meuwissen et al., 2021). This integrative approach has the potential to further improve the accuracy of genomic predictions and facilitate the development of more effective breeding strategies for E. ulmoides and other species (Desta and Ortiz, 2014). 6 Statistical Methods and Tools 6.1 Variable selection techniques The least absolute shrinkage and selection operator (LASSO) is a widely used method for variable selection in QTL mapping and GS. LASSO imposes a penalty on the absolute values of the regression coefficients, effectively shrinking some of them to zero. This results in a sparse model that selects only the most relevant variables. Generalizations of LASSO, such as elastic net and adaptive LASSO, have been developed to address some of its limitations. The elastic net combines the penalties of LASSO and ridge regression, which is particularly useful when dealing with correlated predictors. Adaptive LASSO, on the other hand, assigns different penalties to different coefficients, offering greater flexibility in variable selection (Li and Sillanpää, 2012; Wimmer et al., 2013).
RkJQdWJsaXNoZXIy MjQ4ODYzMg==