Cotton Genomics and Genetics 2025, Vol.16, No.3, 148-162 http://cropscipublisher.com/index.php/cgg 155 In addition to high prediction accuracy, GS can also help break the genetic bottleneck of quality improvement. The "subgenome modular design breeding" proposed by Chinese scholars is an innovative idea: by comparing the differences in the contribution of the two subgenomes of tetraploid upland cotton to fiber development, the genetic modules that restrict quality are identified, and molecular methods are used to recombinantly design the key gene combinations therein, thereby breaking through the limitations of quality improvement. This concept has made progress in preliminary experiments and is expected to be verified in future breeding practices and combined with GS methods (Zhang and Wang, 2024). At the same time, the research of the Xinjiang Production and Construction Corps used the excellent allele variation of sea island cotton to introgress into the upland cotton background, significantly improving the latter's fiber length and strength (Sun et al., 2022). These excellent alleles can also be incorporated into the GS model through marker development to improve the prediction and selection efficiency of quality. In recent years, with the in-depth study of high-quality cotton germplasm resources, several important genes affecting fiber development have been cloned or located, such as GhPAP and other fiber cell wall synthesis-related genes, whose downregulation will lead to a significant decrease in fiber strength. Incorporating these gene loci information into GS can further improve the biological interpretability and effectiveness of the model. At the breeding practice level, India, the United States and other countries have used GS to screen intermediate materials with excellent fiber quality, accelerating the introduction of new lines (Islam et al., 2019). In my country, there are still few studies on genomic prediction of cotton fiber quality, but the relevant foundation has been established: cotton genome sequencing and variation identification have revealed multiple structural variations and candidate genes related to quality; the National Cotton Improvement Center has constructed a series of recombinant inbred line populations with improved quality, which can provide training sets for GS. It can be foreseen that in the near future, breeders will be able to use AI models to pre-evaluate the fiber quality and yield of hundreds of recombinant offspring at the same time, and select excellent individuals that "have both fish and bear's paw". This will greatly improve the efficiency and success rate of breeding of high-quality cotton varieties in my country, and promote fiber quality breeding into a new stage of intelligence. 5 Case Studies: Practical Applications of AI in Cotton Breeding 5.1 Genomic prediction practices in CSIRO's cotton breeding program in Australia The cotton breeding project of CSIRO, Australia is one of the examples of the successful application of genomic prediction technology in crop breeding. Faced with the dual goals of improving fiber quality and maintaining high yield, the breeding team of CSIRO began to try to incorporate GS into its breeding program in the 2010s. They collected multi-season field phenotypic data of thousands of breeding materials and performed high-density genotyping on them. In the process of improving the germplasm material resistant to two-spotted spider mites through backcrossing, despite the continuous advancement of backcrossing generations (BC generations), the mite resistance trait score of the selected material was always significantly better than that of the susceptible parent Sicot 714B3F and remained stable. This resistance stability reflects that CSIRO has effectively retained the target traits through GS and improved the disease and insect resistance without sacrificing the main agronomic traits. The picture shows the significant differences between resistant and susceptible varieties under natural infection at the phenotypic level, providing visual verification for the superior individuals selected by the GS model (Figure 2) (Conaty et al., 2022). Li et al. (2022) reported in detail the results of CSIRO's implementation of genomic prediction on 1,385 lines: the Bayesian LASSO model was combined with genomic SNP and pedigree data to predict traits such as fiber length, strength and yield, achieving remarkable accuracy (length 0.76, strength 0.65, yield 0.64). Of particular note, they found that the fusion of whole genome marker information with conventional pedigree data can effectively improve prediction accuracy, indicating that genomic data has formed a beneficial supplement to traditional breeding information. In practice, CSIRO has used GS to assist in the screening of early-generation materials: for new combinations that have not been field tested, the fiber quality and yield potential are predicted by genotype, and combinations with poor prediction values are eliminated, thereby reducing the workload of field trials and accelerating the generation process. It is reported that after applying GS, the breeding cycle of its new varieties was shortened by about 2 years, and the resource utilization efficiency was significantly improved (Li et al., 2022).
RkJQdWJsaXNoZXIy MjQ4ODYzNA==