ChromSimPipe

Cohesin Loop-Extrusion Simulation Pipeline

Overview

ChromSimPipe is a physics-based polymer simulation framework for testing mechanistic hypotheses about 3D chromatin organisation. Given CTCF binding data (CUT&Tag or ChIP-seq peaks) and Hi-C maps as input, it simulates cohesin loop extrusion under different parameter conditions and compares the resulting contact maps to your experimental Hi-C.

Note

Included example. The default configuration ships with a use case from the STRS project (Flores et al. 2026): HEK293T cells under hyperosmotic stress (sorbitol treatment), comparing control vs. sorbitol CTCF CUT&Tag peaks against matched Hi-C maps. Swap in your own data and the pipeline runs the same way.

How the pieces fit together

  CTCF peaks (CUT&Tag / ChIP-seq)      Experimental Hi-C (.hic maps)
    condition A + condition B              condition A + condition B
          │                                       │
          ▼                                       ▼
  data/ctcf_beds/*.bed             data/mcool/*.cool
   (oriented CTCF sites,            (1 kb contact maps; converted by hic2cool)
    from FIMO JASPAR MA0139.1)
          │                                       │
          └──────────────┬────────────────────────┘
                         ▼
                 configs/parameters.py
           (N conditions × active locus)
                         │
                         ▼
         GPU polymer simulation (polychrom / OpenMM)
         SLURM jobs: N conditions × reps × shards
                         │
                         ▼
              results/polychrom_3d/merged_*/
                         │
                         ▼
              scripts/run_analysis_all.py
           (contact map, P(s), APA, MSD,
            experimental Hi-C comparison)
                         │
                         ▼
                results/analysis/*.npy, *.png

Example conditions (STRS use case)

The default configs/parameters.py defines five conditions for the STRS hyperosmotic-stress example (HEK293T, hg38):

#	Condition	Cohesin params	CTCF sites	Question
1	control_ctcf-control	Gabriele 2022	Control CUT&Tag	Can we reproduce untreated Hi-C?
2	control_ctcf-sorbitol	Gabriele 2022	Sorbitol CUT&Tag	Does CTCF loss alone explain sorbitol Hi-C?
3	weak_ctcf-sorbitol	Gabriele 2022 (0.5× capture)	Sorbitol CUT&Tag	Weaker stalling at retained sites?
4	sorbitol_promoter-stall	Gabriele 2022	Sorbitol + promoter anchors	Promoter-anchored cohesin?
5	sorbitol_promoter-stall_long	Gabriele 2022 (2× processivity)	Sorbitol + promoter	Same + longer processivity?

Example loci (STRS use case, hg38, HEK293T)

Key	Chrom	Window	Genes
chr1_fig1	chr1	64.2 – 66.2 Mb	JAK1 / AK4 / LEPR
chr4_fig1	chr4	1.7 – 3.7 Mb	RNF4 / ADD1 / HTT
chr6_fig1	chr6	124.1 – 126.1 Mb	NKAIN2 / RNF217
chr16_sox8	chr16	0.4 – 2.3 Mb	SOX8

Cohesin model

The simulation models cohesin loop extrusion: cohesin rings load onto chromatin and extrude DNA into growing loops, stalling when they encounter CTCF sites from the correct direction. The default parameter set is derived from Gabriele et al. 2022 (Science) single-molecule imaging in mESCs.

Parameter	Default	Meaning
lifetime	75 steps	~150 kb processivity at 1 kb/monomer
separation	240 monomers	~8 cohesin rings per 2 Mb locus
ctcf_capture	0.125	12.5% stalling probability per CTCF encounter
ctcf_release	0.0033	Stalled cohesin lives ~4× longer than free cohesin

The tiling trick

The 2 Mb locus (~2,000 monomers) is simulated as 28 tiled copies on a 70,000-monomer polymer. This gives 28× more contact statistics per GPU run at no extra physics cost. Contact maps are folded back to 2,000×2,000 for comparison. See Yang et al. 2023 Nat Commun for details.

References

• Flores et al. 2026 — STRS paper. GEO: GSE310051 (Hi-C), GSE310047 (CUT&Tag).
• Fudenberg et al., Cell Reports 15:2038–2049 (2016). Canonical loop-extrusion model.
• Banigan et al., eLife 9:e53558 (2020). LEF dynamics and CTCF barriers.
• Gabriele et al., Science 376:496–501 (2022). Live-cell cohesin imaging (parameter source).
• Yang et al., Nat Commun 14:1913 (2023). Tiling trick for convergence.