Figure 2

When collapsing probes to genes, 1.max is usually the optimal collapsing strategy to choose. A) A typical example of ranked expression (left column) and ranked connectivity (right column) correlation between two data sets. Each dot represents a gene in common between data sets, with the x and y axes represented that gene's ranked expression or connectivity in data sets 1 and 2, respectively. B-D) Across several studies in human brain (B), mouse brain (C), and human blood (D) the MaxMean (1.max) parameter generally produces better ranked expression correlations (left column) than maxVariance (2.var). For both MaxMean and maxVariance, use of connectivityBasedCollapsing (3.kMax and 4.kVar) decreases the between-study correlations. Similar results hold, to a lesser extent, with connectivity correlations (right column). Y-axes correspond to the average expression and connectivity correlation between data sets. Error bars represent standard error. Percentages indicate the percent of assessments in which the relevant strategy had the highest overall between-set correlation.