Fig 1.
Frequency of co-expression is higher than expected by chance in natural transcriptomes.
A and B) Typical co-expression structure and modular organiza-tion of a random sample of 1000 genes from the human brain transcriptome (Brain-span dataset) and 60 brain tissue samples obtained from the Fantom5 dataset. C) and D) Distribution of absolute correlations of all gene pairs in a random sample of 1000 genes compared with the distribution resulting from random permutations of expression data in the same genes. Inset: Bars show the mean (±SEM) number of highly correlated pairs (/R/>0.5) among 1000 independent samples of 1000 genes each compared with the expected mean number of highly correlated pairs (absolute correlation) resulting from random permutations of the gene expression values in the same samples.
Fig 2.
High co-expression naturally emerge in synthetic gene regulatory networks.
A) Co-expression clustering dendrogram based on expression data generat-ed by a synthetic GRN after 1000 time steps (iterations). B) The distribution of the absolute correlation of all synthetic gene pairs in in a simulated network of 1000 genes compared with the distribution resulting from random permutations of the same expression data. Inset: Bars show the mean (±SEM) number of highly corre-lated pairs (/R/>0.5) obtained from 1000 independent GRN simulations compared with the expected mean number of highly correlated pairs (absolute correlation) resulting from random permutations of the gene expression values of the same networks.
Fig 3.
Regulators and direct targets show poorly correlated expression levels in synthetic GRNs.
A-D) Typical scatter graphs showing the level of expression of four independent target genes (Tgt, expressed as log-transformed values) as a function of the level of expression of one direct positive regulator (TF). E) Distribution of correlation values of all individual regulator-target pairs involving only positive regulators. F-I) Typical scatter graphs showing the level of expression of four independent target genes as a function of the level of expression of one direct negative regulator (TF). J) Distribution of correlation values of all individual regulator-target pairs involving only negative regulators.
Fig 4.
Transcriptional regulators in natural transcriptomes are poorly correlated with their individual targets.
(A-C): Scatter graphs showing the level of expression of three out of 400 known targets (Tgt, expressed as log-transformed values) of the transcription factor MYC, as a function of the level of expression of this same transcription factor (TF). D) Distribution of correlations between MYC expression against 400 identified MYC targets. (E-G): Scatter graphs showing the level of expression of three independent targets of the transcription factor PAX2, as a function of the level of expression of this same transcription factor. H) Distribution of correlations between PAX2 expression against 400 identified targets. (I-K): Scatter graphs showing the level of expression of three independent targets of the transcription factor SRY, as a function of the level of expression of this same transcription factor. L) Distribution of correlations between SRY expression against 400 identified targets. Ensembl IDs for each individual target in the scatter graphs are indicated. For each transcription factor, the scatter graph shown correspond to percentiles 25, 50 and 75 of the correlation distribution for each set of targets.
Fig 5.
Correlated pairs of genes share common regulators in synthetic GRN.
A) Typical average topological distance (a measure of regulatory proximity in the GRN) between pairs of highly correlated genes (/R/ > 0.8) compared with the average distance between random pairs of background genes in the synthetic regulatory network. In both cases, the associated p value in a T test was below 10–16. B) Bar chart showing the average proportion (±S.E.M) of pairs of genes sharing a common regulator among either highly correlated pairs (/R/ >0.8) or lowly correlated pairs (/R/ < 0.2) found in 1000 independent simulations.
Fig 6.
Pairs of genes sharing a common regulator tend to be more highly correlated than random background and regulator-target pairs of genes in both synthetic and natural transcriptomes.
A) Average co-expression (absolute correlation) between those pairs of genes sharing a common regulator, as well as regulator-target pairsin the synthetic GRN (arrows) compared with the distribution of average correlations of 1000 equally sized samples of random background pairs of genes. B) Average coexpression (absolute correlation) between pairs of targets of either MYC, PAX2 or SRY, as well as regulator-target pairs involving all these three regulators, (pooled data, arrows) compared with the distribution of average correlations of 1000 equally-sized samples of random background pairs of genes using expression data derived from the Brainspan database.