Selection of appropriate genomic selection model in an unstructured germplasm set of peanut (Arachis hypogaea L.)

Prediction of genomic values with better accuracy is a key to success in genomic selection (GS) breeding. With an objective to deploy GS in peanut breeding, we evaluated six GS models in an unstructured germplasm set containing 184 individuals. In this context, multiseason phenotyping data for three important agronomic traits (days to flower, seed weight and pod yield) and genotyping data with 2356 DArT markers were generated on this germplasm set. Phenotypic analysis revealed lowest coefficient of variation (CV), genetic variance (GV), genotypic coefficient of variation (GCV) for days to flower followed by seed weight and pod yield. In contrast, the heritability was highest in days to flower (78.85%) than seed weight (75.46%) and pod yield (62.53%). Upon analysing phenotypic together with genotypic data with six GS models, cross-validation values were found to be higher for days to flower followed by seed weight and pod yield across all the models. Of the six GS models used, Ridge Regression-BLUP and Bayesian LASSO performed better than Random Forest Regression, Kinship GAUSS, BayesCπ and BayesB. Thus, Ridge Regression-BLUP and Bayesian LASSO may be the choice for further GS analysis in peanut as these two models performed better for high (days to flower and seed weight) as well as moderate (pod yield) heritable traits. Further analysis is continue on this germplasm set and detailed results will be presented in the conference.