| In the present study, methods of constructing rice core collections were employed for 992 rice varieties with 13 quantitative traits. The genotypic values of the traits were predicted by the adjusted unbiased prediction (AUP) method. Based on the genotypic values, the Mahalanobis distances were calculated to measure the genetic similarities among rice varieties. Six hierarchical clustering methods including the completed linkage method, the median linkage method, the centroid method, the unweighted pair-group average method, the flexiable-beta method and the Ward's method, combining with three sampling strategies including of the random sampling, the preferred sampling and the deviation sampling, were proposed to develop 18 core collections of rice germplasm. Four statistical parameters, which were the percentage of significant difference (VD%, a = 0.05) between core collection and initial collection for variance of traits, the percentage of significant difference (MD%, u = 0.05) between core collection and initial collection for means of traits, coincidence rate (CR%) and variable rate (VR%), were applied to evaluate the representation of core collection for initial collection. Main results were gained as follows:1. The comparison of Malialanobis distance and Euclidean distanceThe adjusted unbiased prediction method was used for predicting the genotypic values. On the basis of the predicted genotypic values, the genetic distances among different varieties were calculated by Mahalanobis distance and Euclidean distance. The unweighted pair-group average method, combining with the deviation sampling strategy, was employed to construct two core collections with these two different genetic distances for measuring the genetic similarities among rice varieties. According to the results of the evaluation, the core collections based on Mahalanobis distance could remain more genetic diversities of the initial collection than the ones that based on Euclidean distance.2. Construction of eighteen core collectionsBased on the genotypic values predicted by the adjusted unbiased prediction method, 18 core collections were constructed by 6 cluster methods including the completed linkage method, the median linkage method, the centroid method, the unweighted pair-group average method, the flexiable-beta method and the Ward's method, combined with 3 sampling strategies including the random sampling, the preferred sampling and the deviation sampling. The result showed that MD% for 18 core collections were all lower than 20%. CR% for all core collections' were higher than 80%, except for the core collection constructed by thecentroid method and random sampling strategy. So there were 17 core collections could measure up to the rules of core collections' building.3. Evaluation of the cluster methods and sampling strategiesThe different cluster methods were evaluated under the conditions of the same sampling strategy. When the random sampling strategy was employed, the flexiable-beta method was the best cluster method. When the sampling strategy is preferred, the best cluster way would be the Ward's method. The unweighted pair-group average method would be the best cluster way when deviation sampling strategy was used to develop the core collections. Considering the parameters of the 18 core collections overall, GCoreC4S3 (the unweighted pair-group average cluster method combined with deviation sampling strategy) had the highest VD% and VR%, the relatively high CR%, the lowest MD%, and it would be, therefore, the best one in the 18 core collections. So the deviation sampling strategy, came along with the unweighted pair-group average method of hierarchical clustering, could retain the genetic diversity of the initial rice collections to the greatest degree.4. Comparison of core collections based on genotypic and phenotypic valuesThe best cluster methods, coming along with their corresponding sampling strategies, were selected to compare the difference of core collections based on genotypic and phenotypic... |
PDF Full Text Request