Figure 1.
Study overview.
Table 1.
Genomic distribution of CpG loci in HEP data for each chromosome.
Figure 2.
The workflow of the proposed distance measure.
Table 2.
Significant features in the feature-reduced model, together with corresponding P-value.
Figure 3.
The homogeneity based clustering.
Nine modules with distinct (epi)genomic patterns, involving significant features based on regression analysis. X-axis: the significant epigenomic features from Table 1, excluding histone acetylation marks. The order of features is arbitrary. Y-axis: the closeness of features around CpG loci as measured by CM. Symbol height represents the median CM value. Error bars represents standard error. The number under the Module_i (I = 1, …,9) is the count of CpG loci in Module_i, The specific proportion of each module is shown in Figure 4A. Modules are roughly classified into four meta-groups according to genomic elements and PolII patterns, as indicated by the four colors.
Figure 4.
The loci distribution and function evaluation for all modules.
(A)The proportion of CpG loci in Module_i. (B) The homogeneity in each module as reported by the CLICK algorithm in the EXPANDER package. See the EXPANDER manual for details. The colors of bars are consistent with Figure 3. (C) The GO similarity of each module. The colors are consistent with Figure 3. Only Biological Process terms were used to calculate gene similarity. The similarity values standard deviation of each module is indicated as error bars, with one standard deviation in each direction. (D) The gene expression levels of 9 modules. The gene expression data is probed from resting T cells (M0 and T0 types). The bar height represents the median Log2 values of modules. The standard deviation of the Log2 values of each module is indicated as error bars with one standard deviation in each direction.
Figure 5.
The global Bayesian network inferred from the complete CM profile.
The red, blue and green circles represent different groups. The node color is mapped by the color key in the left top corner, representing the coefficients of the regression model (Table S2).
Figure 6.
Network metrics for the Bayesian network.
Network manipulations were performed in Cytoscape, and the clustering coefficient, average degree, power exponent in degree distribution, and mean shortest path were derived by the Cytoscape Random Networks plug-in.
Figure 7.
The Bayesian network inferred from nine modules.
The node color is mapped by the color key in the left top corner, representing the coefficients of the regression model (Table S2). All node color is mapped by the color key (Figure 5).
Figure 8.
The between-feature interactions from module-specific Bayesian inference.
The top panel shows the occurrence of between-feature interactions in nine Bayesian network maps, where each cell represents one specific interaction type and the number within each cell represents the number of occurrences in all nine modules. The yellow color marks the module-specific and significant interactions discovered at least in two modules. The hatch marks the regulatory relationships only in the global network (Figure 5). The number in the square counts the occurrence of between-feature interactions in nine modules. The bottom panel shows the genomic distribution of each feature.