Figure 4

Performance of SPEC using split subset-specific signatures on gene expression profiles from healthy subjects. Each of the cell subset-specific signatures used in SPEC (individual pie charts) was randomly split 200 times, with half of the genes used as the query signature and the other half retained as the cell subset signature. SPEC was considered to be correct if it linked the query to the subset from which it was derived (shades of blue) and incorrect otherwise (shades of red). Significance cutoffs were calculated for each run to determine whether the link between query and subset was statistically significant (p < 0.05, dark blue/red) or not (p > = 0.05, light blue/pink). Predictions that were incorrect and statistically significant are separated and labeled with the predicted subset. Predictions that did not reach the level of statistical significance are labeled NS+ and NS- to indicate correct and incorrect predictions, respectively. SPEC was applied separately to predict the source of query signatures coming from (A) B cells, T cells, NK cells, neutrophils or monocytes, and (B) lymphoid or myeloid cells.