Figure 1.
A schematic representation of the relationship of the non-reverted and various conditions of the reverted HMT3522 breast cells.
The nonmalignant S1 cell is the root of the tree. It is also the parent of the malignant T4-2 cell since T4-2 cells were derived from S1. The T4-2 cells can be reverted to phenotypically normal-looking structures by treatment with various agents, such as: i) either EGFR or β1-integrin inhibitor, ii) either PI3K or MAPKK inhibitor, or iii) MMP inhibitors, they are thus represented as the parent of the various conditions of the reverted T4-2 cells. Microarray profiles were generated from each cell state represented in the tree, and gene networks specific to each state were reverse engineered using Treegl.
Figure 2.
An example demonstrating the difference between a Markov network and a correlation network.
(A) A true network, in which R is the regulator of A, B, and C. (B) A clique graph produced by a correlation network. Since all the genes in (A) are correlated with one another, the correlation network cannot distinguish between indirect and direct relationships and thus connects all the nodes. (C) A correct graph recovered by a Markov network. The Markov network recovers true relationship of the nodes because it uses conditional independence to determine the presence of an edge.
Figure 3.
Simulation results comparing the performance of Treegl to a method estimating a single static network and a method estimating each network independently.
In all cases, 70 networks were generated that are related by a tree lineage (See Materials and Methods for details). (A) Each network has 30 nodes and 5 samples. (B) Each network has 30 nodes and 10 samples. (C) Each network has 50 nodes and 5 samples. (D) Each network has 50 nodes and 10 samples. In all cases, Treegl (shown in blue) performs favorably to the method estimating a single static network (red) or the method estimating each network independently (green).
Figure 4.
Illustration of selected enriched pathways in the differential network of each breast cell state.
(A) S1; (B) T4; (C) the EGFR/ITGB1-T4R group; (D) the PI3K/MAPKK-T4R group; and (E) the MMP-T4R group. In each plot, the differential network for the corresponding cell state is shown. Nodes represent genes; edges represent interaction of the genes. Selected pathways enriched in the network (unadjusted p-values<0.05) are color-coded and shown on the right; genes participating in the selected pathways are also colored based on the color code for the corresponding pathway. An asterisk indicates a pathway that is significantly enriched in the corresponding differential network. The pathway names are shortened to save space. See Table S1 for detailed information about the enriched pathways in each cell state.
Figure 5.
Diseases associated with the genes in the differential network of each breast cell state.
(A) S1; (B) T4; (C) the EGFR/ITGB1-T4R group; (D) the PI3K/MAPKK-T4R group; and (E) the MMP-T4R group. In each plot, the differential network for the corresponding cell state is shown. Nodes represent genes; edges represent interaction of the genes. Diseases associated with the genes in the network (unadjusted p-values<0.05) are color-coded and shown on the right; genes associated with a certain disease are also colored based on the color code for the corresponding disease. An asterisk indicates a disease that is significantly enriched in the corresponding differential network.
Figure 6.
Kaplan–Meier curves estimating the association of different expression values of three hubs in the differential networks of the breast cell states with survival of breast cancer patients.
(A) NEBL in the S1 differential network; (B) HBEGF in the T4-2 differential network; (C) PAPD7 in the MMP-T4R differential network.
Figure 7.
Selected pathways or GO groups in the neighborhood of the three hubs in the differential networks of the breast cell states.
(A) NEBL in the S1 differential network; (B) HBEGF in the T4-2 differential network; and (C) PAPD7 in the MMP-T4R differential network. An asterisk indicates a pathway or a GO group that is significantly enriched in the corresponding differential network.