**9. Genomic selection in cotton**

QTL mapping and genome-wide association studies have identified many genomic regions responsible for the important agronomic and fiber quality traits in cotton. Among them only a few traits like disease resistance and pest resistance were qualitatively governed by a few genes/QTLs with a major effect. Marker-assisted selection (MAS) is well-suited for handling these traits. But in the majority of the crops and also cotton, most of the yield, yield contributing and fiber quality traits are quantitatively governed by one or few QTLs with relatively large effects along with several QTLs with small effects, which are not captured through QTL mapping [128, 129]. Hence targeted phenotype has not been achieved successfully through Marker Assisted Selection. Under such a situation, genomic selection (GS) would seem to be a promising and powerful tool of genomics to breed for these traits. GS is a unique form of MAS, here the basis of selection is the genotypic data on marker alleles covering the entire genome, irrespective of whether the effects associated with these marker loci are significant or not [130]. Based on these marker effect estimates, genomic estimated breeding values (GEBVs) of different individuals/lines will be calculated without actually phenotyping them, which forms the basis of selection (**Figure 7**). GS empirical studies in maize (*Zea mays*; [132–135]), rice (*Oryza sativa*; [136–139]), wheat (*Triticum aestivum*; [140–144]), and sorghum (*Sorghum bicolor*; [145–147]) have all recently shown how GS has become an efficient approach in crop breeding with recent developments in the implementation of various high-density array-based DNA marker technologies and their reduced genotyping costs. There are many marker effects estimation models that have been developed for the GS. Their predictability mainly depends on factors like marker density, training population size, and the relationship between training and breeding populations [131, 148]. Hence, the model which is capable of giving the highest GEBV accuracy will be selected. To date only two cotton GS studies have been reported. Islam et al. (2020) compared prediction ability (PA) and prediction accuracy (PACC) of Several GS models in cotton including genomic BLUP (GBLUP), ridge regression BLUP (rrBLUP), BayesB,

#### **Figure 7.** *Schematic representation of genomic selection (GS) scheme (based on [131]).*

Bayesian LASSO, and reproducing kernel Hilbert spaces (RKHS). And reported BayesB predicted the highest accuracies among the five GS methods tested and also the same model is suggested by Gapare et al. in 2018 in cotton. In many field crops for different traits, GS prediction accuracies of >0.80 have been reached [149, 150], but now in cotton, the accuracy of 0.71 and 0.59 for fiber length and strength has been achieved, respectively [150]. The prediction ability (PA) and prediction accuracy (PACC) was 0.65, and 0.69, respectively for fiber elongation [148]. In most plant breeding programs, especially in cotton, GS is still in its infancy and one of the biggest barriers to the implementation of GS in practical plant breeding is the high start-up cost required for accurate phenotyping, maintaining a large training population and costs of genotyping entire breeding populations. However, nowadays the genotyping costs are continually decreasing and genotyping of large plant populations is much more manageable than going in for conventional phenotyping. Soon, at points in the breeding program where selection using conventional methods is too costly and time-consuming, GS may have its greatest potential usage.
