collapseRows accurately predicts the relative quantity of cell type across samples of whole blood. Using data from a realistic blood model (Grigoryev et al 2010), the 1.max, 3.kMax, 5.ME, and 6.Avg collapseRows aggregation strategies can still predict the relative proportion of several major cell types. Each point represents the correlation between true and predicted proportions for one of the four strategies. The X-axis corresponds to the number of marker genes used for the predictor, while the Y-axis corresponds to the correlation between true and predicted proportions. Note that all methods other than MaxMean (1.max) are relatively robust to choice in number of marker genes.