Fig 1.
An illustration of the multi-scale chromatin organization.
The approximate genome lengths (in base pairs) are presented alongside structural models and experimental contact maps for each scale (data: 4DN Data Portal 4DNES4AABNEZ - in situ Hi-C on human embryonic stem cells (H1) treated with RNase A). Abbreviation: TAD - topologically associated domain.
Fig 2.
An overview of the strategies for chromatin structure modeling.
Strategies for chromatin modeling can be divided into genomic data-driven, image-driven and predictive. Each strategy can produce a structural 3D model or a contact map.
Table 1.
Overview of software for chromatin structure modeling for genomic data-driven strategy. Software for chromatin modeling can be categorized into different categories based on input data, modeling scale or output format.
Table 2.
Examples of software for chromatin structure modeling made with other than genomic data-driven strategy. Software for chromatin modeling can be categorized into different categories based on input data, modeling scale, modeling strategy or output format.
Fig 3.
Workflow for benchmarking of chromatin structural models.
The workflow developed at the 4D Nucleome Hackathon 2024 consists of multiple steps that include obtaining, visualizing, processing and comparing models of chromatin structure.
Fig 4.
Chromatin models based on HiC data showing the same topologically associated domain.
Models presenting the same genomic region (chr1:178.421.513–179.491.193) obtained from five software packages (DIMES, MultiMM, MiChroM, LoopSage, and PHi-C2).
Fig 5.
Chromatin models based on ChIA-PET data showing the same topologically associated domain.
Models presenting the same genomic region (chr1:178.421.513–179.491.193) obtained from five software packages (DIMES, MultiMM, MiChroM, LoopSage, and PHi-C2).
Table 3.
Comparison of chromatin models generated based on Hi-C data. The table contains Spearman correlation coefficients calculated between pairs of distance matrices generated based on the models from five software packages (DIMES, MultiMM, MiChroM, LoopSage, and PHi-C2).
Table 4.
Comparison of chromatin models generated based on ChIA-PET data. The table contains Spearman correlation coefficients calculated between pairs of distance matrices generated based on the models from five software packages (DIMES, MultiMM, MiChroM, LoopSage, and PHi-C2).
Fig 6.
Comparisons of heatmaps generated from Hi-C models with a) the highest correlation (DIMES vs MultiMM), and b) the lowest correlation (PHi-C2 vs MiChroM).
All four heatmaps correspond to the interpolated models (number of beads = 214) that were generated for the same region of interest (chr1:178.421.513–179.491.193) based on Hi-C data.
Fig 7.
Spearman correlation between simulated and experimental data.
The heatmaps display correlations for models generated from ChIA-PET and Hi-C data. Both types of models show stronger agreement with Hi-C experimental data, possibly due to the smoother structure of Hi-C-based models, which aligns more closely with experimental observations.
Fig 8.
KL divergence between simulated and experimental data.
The heatmaps show results for models generated from ChIA-PET and Hi-C data. Consistent with the Spearman correlation analysis, both model types exhibit better agreement with Hi-C experimental data.
Fig 9.
Stratified Spearman correlations between simulated and experimental data.
Most models perform poorly in strata smaller than the resolution used to generate the final model. Typically, a simulation model shows good agreement only at the resolution it was designed to represent.