Figure 2

Comparing CutTree with a random search. (a) The structure of the gene network affects the performance of CutTree. A more treelike structure improves its performance. A low depth indicates a shallow tree with many branches; a higher depth indicates a tree with few branches and long paths. Simulation data from 13 in silico gene networks with different structures, each being affected by a single PAG. Mean values from 50 experiments are shown (see Methods). (b) In silico simulations of the mean number of experiments required to find one PAG among various numbers of differentially expressed genes. For one PAG and a few genes, CutTree performs approximately equal to a random search but performs substantially better as the number of genes increases. For two PAGs, CutTree clearly outperforms a random search. (c) Schematic of gene networks with the PAG at the top.