We tackle gene expression estimation from pathology images by proposing STRank, a novel loss function that captures relative relationships of expression from multiple samples. This approach improves robustness to batch effects and stochastic noise.