Fig 1.
Overview of the MaxComp algorithm working on single-cell structures
(A) The assumption of chromatin compartmentalization is based on two parts: Firstly, chromatin regions belong to the same compartment have higher contacts than those from the different compartment; Secondly, chromatin regions of compartment A are spatially closer to nuclear speckles than regions of compartment B. (B) Every single-cell structure is transformed to an undirected graph which can be represented by an adjacency matrix whose edge weights are decided by its pairwise distances and speckle distances. Max Cut is then applied to the matrix to generate two partitions of nodes which are the prediction of the single-cell compartments of the structure. Ensemble compartment frequency can be calculated by combining a population of single-cell profiles.
Fig 2.
Selected example of model compartments predicted by MaxComp and the corresponding structures of H1-hESC Chr6
(A) The predicted compartments for structure 35, 115, 255, 361, 443 of H1-hESC Chr6 together with the input adjacency matrices of the MaxComp approach. (B) The compartment profile and 3D visualization of structure 35, 115, 255, 361, 443 of H1-hESC Chr6 colored by compartments (red in compartment A and blue in compartment B) from both experiment and the MaxComp prediction showed together with the nucleus envelope. (C) Predicted speckles (green) showed together with the single-cell examples colored compartments labeled by MaxComp and the Hi-C-based PCA.
Fig 3.
Prediction of model compartments by MaxComp and its comparison with other methods on H1-hESC Chr6 and Chr10
(A) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 500 modeled structures of H1-hESC Chr6. (B) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 500 modeled structures of H1-hESC Chr10. (C) Scatter plot between the normalized compartment frequencies of MaxComp and the PC1 values together with the Pearson’s correlation coefficient between the two samples on each chromosome. (D) Scatter plots of MaxComp single-cell compartment variabilities against PC1 values from the Hi-C-based PCA annotations for each chromosome. Each dot represents a genomic region colored by its value of cell-to-cell radial position variability in their radial positions (Left) or cell-to-cell speckle distance variability (Right). (E) Illustration of A-associated locus and B-associated locus acting as anchors at speckles and envelope during chromatin folding. (F) Comparison of compartment profile (p-value = 8.84e-139 and 4.74e-140), compartment variability (p-value>0.05), radial position variability (p-value>0.01) and speckle distance variability (p-value = 4.36e-59 and 5.21e-61) between top 20% PC1 locus and bottom 20% PC1 locus of H1-hESC Chr6 and Chr10. (G) Comparison of speckle distances (p-value = 1.37e-55 and 3.64e-152), lamina distances (p-value = 3.24e-63 and 1.57e-60), radial position (p-value = 3.24e-63 and 1.57e-60), radius of gyration (p-value = 3.12e-44 and 4.71e-50) and interchromosomal contacts (p-value = 2.87e-53 and 3.02e-73) between compartment A beads and compartment B beads on 500 structures of H1-hESC Chr6 and Chr10. (H) Comparison of compartmentalization scores between the Hi-C-based PCA annotation, the MaxComp prediction, the distance-based PCA annotation and the CG phasing prediction for each structure from H1-hESC Chr6 and Chr10 showed together with the corresponding violin plots and illustration.
Fig 4.
Prediction of DNA-MERFISH compartments by MaxComp and its comparison with other methods on IMR90 Chr6 and Chr10
(A) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 7,000 DNA-MERFISH structures [7] of IMR90 Chr6. (B) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 7,000 DNA-MERFISH structures of IMR90 Chr10. (C) Scatter plot between the normalized compartment frequencies of MaxComp and the PC1 values together with the Pearson’s correlation coefficient between the two samples on each chromosome. (D) Scatter plot of compartment variabilities against PC1 values from the Hi-C-based PCA annotations on each chromosome. Each dot represents a genomic region colored by its value of speckle distance variability. (E) Comparison of speckle distances (p-value = 0.0 and 0.0), lamina distances (p-value = 1.58e-102 and 2.20e-119) and numbers of imaged nascent transcripts (p-value = 6.23e-06 and 4.76e-60) between compartment A beads and compartment B beads on 7,000 DNA-MERFISH structures of IMR90 Chr6 and Chr10. (F) Comparison of compartmentalization scores between the Hi-C-based PCA annotation, the MaxComp prediction, the distance-based PCA annotation and the CG phasing prediction for each structure on IMR90 Chr6 and Chr10 showed together with the corresponding violin plots. (G) The 3D visualization of 6 selected DNA-MERFISH structures of IMR90 Chr6 colored by compartments (red in compartment A and blue in compartment B) from both ensemble PC1 and Max-cut prediction.
Fig 5.
+ compartments by MaxComp and its comparison with other methods on mESC Chr5 and Chr15 (A) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 886 SeqFISH+ structures [8] of mESC Chr5. (B) The experimental profile obtained from the Hi-C-based principal component analysis and the compartment profile predicted by MaxComp on 884 SeqFISH+ structures of mESC Chr15. (C) Scatter plot between the normalized compartment frequencies of MaxComp and the PC1 values together with the Pearson’s correlation coefficient between the two samples on each chromosome. (D) Scatter plot of compartment variabilities against PC1 values from the Hi-C-based PCA annotations on each chromosome. Each dot represents a genomic region colored by its value of speckle distance variability. (E) Comparison of compartmentalization scores between the Hi-C-based PCA annotation, the MaxComp prediction, the distance-based PCA annotation and the CG phasing prediction for each structure on mESC Chr5 and Chr15 showed together with the corresponding violin plots. (F) The 3D visualization of 6 selected SeqFISH+ structures of mESC Chr5 colored by compartments (red in compartment A and blue in compartment B) from both experiment and the MaxComp prediction. (G) Log fold change of the average transcription level (number of mRNA transcript spots detected) of cells with both copies labeled with A (A cells) against the level of cells with both copies labeled with B (B cells) for each gene.