Fig 1.
Cross validations performed to estimate the effect of population structure in genomic selection.
A. 2nd cross validation strategy–Within-cluster cross validation formed by discriminant analysis of principal components (DAPC); B. Illustration of between-clusters cross validation for DAPC cluster 1: 3rd cross-validation strategy–cross validation with training population composed of a third part of DAPC clusters 2–5 and validation in a third part of DAPC cluster 1 (1stTP); 4th cross-validation strategy–cross validation with the training population composed by a third part of three of the DAPC clusters (2nd, 3rd 4th, and 5th TPs). Therefore, the first training population (1st TP) is formed from the four remaining DAPC clusters (3rd cross validation strategy), while the second, third, fourth and fifth training populations (2nd, 3rd 4th and 5th TPs) were each created from only three of those four DAPC clusters (4th cross validation strategy).
Table 1.
A posteriori means for genomic estimates of fresh root yield, dry root yield and dry matter content for the first cross-validation strategy.
Fig 2.
Predictive ability and bias boxplot for different genomic selection methods (BayesB, BLASSO, RR-BLUP, G-BLUP and RKHS) with the 1st cross validation strategy, for fresh root yield, dry root yield and dry matter content.
Table 2.
LRT analysis and Tukey’s pairwise test (p≤0.05) for genomic selection prediction for the first cross-validation strategy for fresh root yield, dry root yield and dry matter content.
Fig 3.
Pearson correlations of genomic estimated breeding values (GEBVs) from different genomic selection methods (BayesB, G-BLUP, RKHS, RR-BLUP and BLASSO) with the 1st cross validation strategy for fresh root yield, dry root yield and dry matter content.
Fig 4.
Cohen’s Kappa for the selection coincidence of higher GEBV cassava genotypes, considering selection proportion (5% to 30%—SP) with the 1st cross validation strategy of different genomic selection methods BayesB, BLASSO, G-BLUP, RR-BLUP, RKHS and phenotypic REML/BLUP, for fresh root yield, dry root yield and dry matter content.
Fig 5.
Distribution of the five clusters identified by discriminant analysis of principal components of single nucleotide polymorphism markers (SNPs).
Table 3.
LRT analysis and Tukey’s pairwise test (p≤0.05) for genomic selection prediction considering the 2nd, 3rd, and 4th cross-validation strategies to estimate the effect of population structure for fresh root yield, dry root yield and dry matter content.