Fig. 3. CONTENT outperforms existing approaches in the GTEx and scRNA-seq CLUES datasets.

A, D Number of genes with a significantly predictable component (hFDR ≤ 5%) in GTEx (A) and CLUES (D); the sample sizes for each context are included in parentheses. B, E Ratio of expression prediction accuracy (adjusted R²) of the best-performing cross-validated CONTENT model over the context-by-context (green) and UTMOST (blue) approaches (median across all genes significantly predicted by at least either method). Numbers above one indicate higher adjusted R² and thus prediction accuracy for CONTENT. C, F Prediction accuracy of CONTENT(Full) and CONTENT(Shared) when a gene-tissue has a significant shared, specific, and full model.