Fig. 3
From: A framework for transcriptome-wide association studies in breast cancer in diverse study populations
Predictive performance of key genes, accounting for sampling variability. Validation R2 across PAM50 molecular subtype and estrogen receptor status, stratified by race, for example genes with highly variable R2 in TCGA (a) and held-out CBCS (b). Squared Spearman correlation (Y-axis), denoted R2, between observed and predicted gene expression is plotted for different genes (X-axis), stratified by PAM50 subtype and estrogen receptor status. Points are colored and shaped according to subtype. Error bars provide 90% confidence intervals inverted from the corresponding permutation test