Fig 1.
Association between chromatin interaction and co-expression of gene-pairs.
(A) Co-regulated genes are co-localized in 3D structure of the genome through frequent chromatin interactions. Chromatin interactions can be at different levels from direct interaction between genes (interaction between Gene A and Gene B) to chromatin compartment interactions (interaction between Gene D and Gene B). Chromatin interactions between gene-pairs can be characterized by a network, termed Chromatin Interaction Network (CIN). (B) Co-expression between gene-pairs based on their spatial expression pattern across the mouse cortex. Gene A, Gene B and Gene D are expressed in the mid layers of the mouse cortex and are hence highly co-expressed. Gene C, on the other hand, is expressed in the superficial cortical layers and therefore is not co-expressed with the other three genes. The chromatin interaction profile of a gene-pair, encoded by the topological structure of the CIN, can be used to predict the co-expression status as captured by the co-expression network.
Table 1.
Topological measures.
Fig 2.
Co-expressed genes are co-localized in 3D structure of the genome.
(A) Assessment of the enrichment of Hi-C interactions between strongly co-expressed gene-pairs compared to gene-pairs with no co-expression across different Hi-C resolutions. The y-axis indicates −log10(p − value) of the one-tailed Wilcoxon rank-sum test used for the enrichment analysis. Hi-C interactions were mapped to genes using the MAX-mapping method. (B) Overview of the Hi-C resolution at which Hi-C interactions are most significantly associated with co-expressed gene-pairs for each chromosome. In each box, the horizontal line represents the median. The thick vertical line represents the interval of q1 = 25th and q3 = 75th percentiles. The thin vertical line represents the interval of q3 + 1.5(q3 − q1) and q1 − 1.5(q3 − q1).
Fig 3.
Chromatin interactions of gene-pairs in the CIN at 200kb resolution.
(A) Synj1-Dykr1a (yellow nodes) in Chromosome 16 are co-expressed (dashed red link) but their corresponding genomic loci do not interact frequently (no blue link). Both genes have strong chromatin interactions with 4 other genes (grey nodes) resulting a high Jaccard index between them. (B) Kcnc4-Tapan5 (yellow nodes) in Chromosome 3 directly interact (solid blue line) but they are not strongly co-expressed (no dashed red line). This direct chromatin interaction explains the strong co-expression between other gene-pairs in their neighbourhood, such as Wdr47-Tapan5 and Wdr47-Rap1gds1, which are not directly connected in the CIN themselves (no solid blue line). The betweenness centrality measure of the link between Kcnc4-Tapan5 can describe the strong co-expression between their neighbouring genes. Chromatin interaction and co-expression are shown by solid blue and dashed red links, respectively.
Fig 4.
Classification performance for the co-expression prediction based on intra-chromosomal chromatin interaction networks.
Each box encompasses the classifier performance in terms of AUC for all mouse chromosomes. Boxes are sorted based on their medians. The method that was used for computing the input feature set is given under each box. TSS or MAX refers to the mapping method for assigning Hi-C interaction between gene-pairs when the CIN is built. TSS + MAX refers to concatenated feature set of topological measures of CINs built using both TSS- and MAX-mapping methods. Multi-resolution refers to concatenated feature set of topological measures obtained from CINs at Hi-C resolutions of 40, 80, 120, 160, and 200kb. In each box, the horizontal line represents the median. The thick vertical line represents the interval of q1 = 25th and q3 = 75th percentiles. The thin vertical line represents the interval of q3 + 1.5(q3 − q1) and q1 − 1.5(q3 − q1). All the values shown in the figure are also available in S1 Table
Fig 5.
Topological signature of Chromosome 16 CIN.
t-SNE maps of the 200kb-CIN of Chromosome 16. Each point in the map corresponds to a link between a gene-pair which is colored according to (A) Clustering coefficient at small-scale, (B) Co-expression label, and (C) Jaccard index at the medium-scale. (D) and (E) show sub-networks of the CIN surrounding selected gene-pairs (indicated in the 2D maps): (D) B3galt5-Carhsp1 (yellow nodes) with the high clustering coefficient average; and (E) Masp1 and Abat (yellow nodes) with the high Jaccard index. Chromatin interaction and co-expression are shown by solid blue and dashed red links, respectively.