Table 1.
Proportion of resampling runs where BLUP using trait-specific genomic relationship matrices (TGRM-BLUP) outperformed genomic BLUP (gBLUP).
| Method | C14:0 | C16:0 | C16:1 | C18:0 | C18:1 | C18:2 | C18:3 | C20:0 | C20:1 |
|---|---|---|---|---|---|---|---|---|---|
| BRR | 0.96 | 1.00 | 0.92 | 1.00 | 0.48 | 1.00 | 0.62 | 1.00 | 0.68 |
| Bayes A | 0.82 | 1.00 | 0.80 | 1.00 | 0.38 | 0.98 | 0.28 | 1.00 | 0.54 |
| Bayes B | 1.00 | 1.00 | 0.96 | 1.00 | 0.54 | 1.00 | 0.58 | 1.00 | 0.92 |
| Bayes Cπ | 1.00 | 1.00 | 0.96 | 1.00 | 0.58 | 0.98 | 0.62 | 1.00 | 0.86 |
| BL | 0.74 | 1.00 | 0.94 | 1.00 | 0.52 | 0.98 | 0.50 | 1.00 | 0.74 |
Marker effects were estimated using five Bayesian whole-genome regression approaches for each of the nine fatty acid traits in the Diversity Panel (336 lines). Predicted marker effects were used to construct TGRMs for each trait. The predictive ability of TGRM-BLUP was assessed using nine fatty acid phenotypes measured in a population of 213 oat lines (Elite Panel). Five-fold cross validation was performed with 50 independent resampling runs. TGRM-BLUP was deemed to significantly improve genomic predictions in a TGRM-BLUP approach that outperformed gBLUP in 90% or more of the resampling runs, and are indicated by boldfaced text. BRR, Bayesian ridge regression; BL, Bayesian LASSO.