Figure 3.

Biological priors based on top GO cellular components improving genomic prediction accuracy for growth related traits. Prediction accuracy of the best GFBLUP models using Gene Ontology (GO) cellular components gene groups, compared to the GBLUP benchmark (without prior biological knowledge). The accuracy of benchmark model may differ within corresponding figures of Figures 2, 3, because it is calculated from the corresponding time point, where improvement by GFBLUP was observed. Accuracy was calculated as Pearson correlation between true vs. predicted values. The GBLUP-ALL model uses all markers in GBLUP; GFBLUP-GO models use the top GO cellular component terms mentioned in the text above. The text in the bottom of boxplots shows the likelihood ratio test p-value (LRT) and proportion of genomic heritability explained (h${}_{\text{f}}^2$) by corresponding GO model. (A) Accuracy for Φ_PSII under low light irradiance levels (LL). (B) Accuracy for Φ_PSII under high light irradiance (HL). Similar to Figure 2, the GFBLUP-GO model did not pass all of our model evaluation criteria (see Model Performance Evaluation), though showing some improvement. (C) Accuracy for PLA.