Figure 5.

Prediction accuracies for two multi-kernel trait-specific BLUP models (MK-TGRM-BLUP) that use TGRM for the three most abundant and three least abundant fatty acids. Estimated marker effects for the most abundant fatty acids (C16:0, C18:1, and C18:2) were used to create the three TGRM and were used in a multi-kernel gBLUP framework to predict total lipid content. A similar approach was used with the three least abundant fatty acids (C14:0, C16:1, and C20:0). Predictions for each MK-TGRM-BLUP approach were compared to a multi-trait gBLUP approach. Five-fold cross validation was performed using 50 resampling runs. BL, Bayesian LASSO; BRR, Bayesian ridge regression; r, Pearson's correlation coefficient.