Cotton Genomics and Genetics 2025, Vol.16, No.5, 222-231 http://cropscipublisher.com/index.php/cgg 228 7 Challenges and Limitations in Multi-Trait GWAS Applications 7.1 Population structure, environmental noise, and phenotype accuracy When analyzed by MT-GWAS, the interference brought by the population structure is often more serious than expected. Especially when the sample itself is unbalanced, population stratification is prone to false positives, especially for data of non-European ancestry or mixed groups, the probability of error is even higher (Troubat et al., 2024). On the other hand, the environment is not an easy factor either. Even a slight fluctuation in climate or human error in field experiments may cause the measurement of traits to deviate from the right direction. Especially in crop research, environmental effects sometimes even outweigh genetic signals, making the estimation of heritability less accurate (Zhang et al., 2019). It would be best if phenotypic data could be accurate and uniform across different environments and groups. However, in practice, it often encounters limitations in terms of resources, human resources and technology, making it difficult to truly achieve large-scale implementation. 7.2 Statistical limitations: power, computational complexity, and false discovery rate Theoretically speaking, MT-GWAS can indeed detect signals more easily than the single-trait method. However, the problem lies in that it is overly data-dependent-the correlation between traits, the genetic background of the population, and the sample size-if these variables do not cooperate properly, the detection capability will immediately decline (Suzuki et al., 2024). To find small effect sites or pleiotropic regions, the sample size needs to be astonishingly large. However, in actual crop research, samples of this scale are not common. Moreover, once the number of traits and SNPS is too large, the pressure on model calculation also increases accordingly. Multicollinearity often makes the results difficult to interpret and may also slow down the analysis efficiency (Porter and O 'Reilly, 2017). Let's talk about multiple tests. Although corrections like Bonferroni are rigorous, they can also easily "overfilter out" some genuine signals. Especially when the experimental error is large, how to balance the detection efficacy and control false positives has become a dilemma. 7.3 Translational gaps between discovery and deployment in breeding programs Even if you identify a bunch of loci, there might not be that many that can actually be put to use. This is a common gap in MT-GWAS applications. Some sites have inherently small effects or are highly sensitive to the environment, resulting in unstable performance in the field. As a result, marker-assisted selection is unlikely to have practical significance. Another rather awkward aspect is that many candidate genes lack functional verification. The combination with other omics data is also often absent (Zhu and Luo, 2024). As a result, you have statistical associations but cannot quickly turn them into practical tools (Khatiwada et al., 2023). Traits themselves are complex enough. Coupled with the interaction between genes and the environment, even beneficial alleles may have vastly different expression results in different locations and years. Therefore, this transformation from "discovery" to "usability" remains the most challenging nut to crack in current breeding. 8 Future Directions and Concluding Remarks Whether MT-GWAS can truly play a greater role in the future ultimately depends on whether the phenotypic data can keep up. Manual measurement alone is no longer sufficient. Technologies such as high-throughput imaging, remote sensing, and automated sensors are increasingly being used to collect field traits-not only are they highly efficient, but they also minimize human errors and enhance the accuracy of the data. However, data alone is not enough. If there is no appropriate algorithm to parse, no matter how much data there is, it will just be piled up. Nowadays, some studies have begun to attempt to incorporate deep learning and artificial intelligence methods to explore the non-linear relationships between traits that are not so easily detectable. This is very helpful for identifying QTLS with small effects and complex genetic backgrounds. Returning to the genomic level, the traditional approach of only looking at a single reference genome has become increasingly difficult to meet the demands. Pan-genome and haplotype analyses are increasingly being adopted by more people. They can not only reveal previously overlooked structural variations and rare alleles, but also enable more precise localization, especially for those variations related to pleiotropic traits. Especially in some
RkJQdWJsaXNoZXIy MjQ4ODYzNA==