Building reproducible 3D genome simulations frameworks by populating the polymer model zoo

Team 9

Geoff Fudenberg�USC

Maxime�Tortora�USC

open2C

Aleksandra Galitsyna�MIT

Euxhen Hasanaj�CMU

Fabiana�Patalano�University of Oslo

Krzysztof Banecki�WUT

Introduction

• 3D genome polymer simulations hold significant importance in understanding chromatin dynamics and Hi-C data interpretation
• Typically, simulation software presents a steep learning curve, requiring knowledge of physics, dealing with complex programming interfaces, and enduring extensive processing times.

​

• Our goal is to develop a software that is easy-to-use and accessible, thereby streamlining and enhancing the toolset available to researchers in the 3D genome domain:
• Python-based API
• User-Friendly
• Support for a variety of forces to accurately model chromatin dynamics
• Specialized in chromatin beads-on-strings model

Polychrom: OpenMM and HOOMD

Project outline:

Performance of different Polychroms

HOOMD is faster than OpenMM at large system sizes

DPD is the fastest mode

Benchmark of Polychroms

<<

<

Radius of Gyration (Rg): DPD vs Langevin

DPD overshoots but converges to the same values as Langevin:

1e4 sim. rounds

Rg

Scaling laws reproducibility: DPD vs Langevin

DPD does not favor long-range interactions, but not drastically

Real Hi-C data

Langevin

DPD

Benchmark of Polychroms

<<

<

Biological properties: compartments

Compartments in cohesin-degron Hi-C data:

DPD HOOMD

Biological properties: compartments

Monomer repeats: 1

Iterations 10⁶

Monomer repeats: 10

Iterations 10⁶

Cohesin-degron Hi-C data

Simulations�naive

Simulations�improved

Benchmark of Polychroms

<<

<

Single-cell Hi-C is an emergent assay in 4DN Consortium

sciHi-C

snHi-C

scHi-C

Single-cell Hi-C: modelling of the data is a complicated problem

sc-polychrom: scHi-C modelling from scratch

Prototype of scHi-C 3D-modelling pipeline:

​

• HOOMD-based
• User-friendly API, easy to customize
• Inspired by works of Longzhi Tan and Tim Stevens�

Features:

• Hierarchical modelling scheme�(from large resolution to small)
• Model interpolation to smaller resolutions
• Contact phasing and resolution

scHi-C pairs

Bin�Resolve by haplotype

HOOMD�3D-modelling

Visualisation

3D-coordinates

Ambiguous pairs

Resolved pairs

Resolve ambiguous

sc-polychrom: accessible single-cell Hi-C simulations

Features:�

• 3D model reconstruction for individual cells
• Whole genome modelling
• Resolving homologous chromosomes

​

​

chr1-paternal

chr1-maternal

chr1-paternal

chr1-maternal

chr1-maternal

chr1-paternal

Imputed haplotype-resolved scHi-C:

Benchmark of Polychroms

<<

<

Benchmarking Pipeline: get started with a few lines of code

Introduced a new benchmarking Python API: polychrom-HOOMD

• Easy to use
• Supports selecting different initializations, and integrator algorithms
• Automatically constructs force fields
• Supports live tracking of the simulation via wandb
• Can log metrics (Rg) as well as images (system snapshots)
• Easy to track GPU performance to uncover bottlenecks in your code

Live tracking of simulations

Simulation integration with higlass/resgen

Monomer repeats: 1

Iterations 10⁶

Monomer repeats: 10

Iterations 10⁶

Results

• Polychrom-HOOMD is making scHi-C simulations accessible and user-friendly.
• Interactive live demonstrations showcasing tracking metrics in polymer simulations, such as execution time, radius of gyration (Rg), and 3D structures.
• Streamlined enhancements to the polychrom-hoomd API for better usability.
• Analysis and comparison across different polychrom variants.
• A structured approach for benchmarking and parallel processing of simulation tasks (snakemake-based).
• Integration of polychrom APIs with other tools from the 4D Nucleome project, like cooler and cooltools.
• Simplification and accessibility of the single-cell Hi-C (scHi-C) modeling algorithms.

https://github.com/4dn-hack-team9-2024

Thank you!

Additional slides

Bonds in single-cell Hi-C data: 3D distance

Langevin vs DPD