Figure 4

collapseRows accurately predicts the relative quantity of blood cell lines across mixed samples. Four prediction methods were used on data from (Abbas et al 2009), from which both gene expression data and actual blood cell counts were known: A) maximum mean expression (1.max), B) maximum connectivity (3.kMax), C) module eigengene (5.ME), and D) average (6.Avg) expression of all marker genes. Each dot presents one cell type in one sample. The X-axes correspond to the predicted proportion of each cell type, while the Y-axes correspond to the actual proportion of each cell type across samples. Values are scaled so that the sum of the proportions for a single cell type across all samples is 1. For all methods (except ME), the x = y line (representing perfect agreement) is plotted. Note that choosing the gene with the highest connectivity (B) most accurately predicts the true cell type proportions.