This GitHub repository provides tools for training machine learning models to predict 3D chromatin architecture from single-molecule footprinting data (e.g., methylation patterns). It integrates sequence features and occupancy profiles to infer genome folding and validates predictions by comparing Hi-C data with simulated chromatin loop extrusion, incorporating locus-specific occupancy rates and dynamic CTCF barriers.
The structure of this repository follows as below:
- output : files after processing and analyzing the input data.
- analysis: notebooks and code for analyzing simulations and experimental data.
- utils: necessary functions and tools for performing workflow
- models: machine learning models for predicting CTCF occupancy rate.
- Polychrom: A toolkit for polymer simulations. (https://github.com/open2c/polychrom)
- OpenMM: A library for molecular simulations. (https://github.com/openmm/openmm)
- Open2C analysis packages (see https://github.com/open2c)
First,
git clone https://github.com/Fudenberg-Research-Group/OccupancyInputCTCF.git
- One-Dimensional Lattice Simulation: with running
workflow.py
After running the simulations, the simulated trajectories can be processed to generate in silico ChIP-seq profiles, 1d contact maps, and 3d contact maps (optional). Scripts for data processing available in processing. Instructions are provided with the relevant python code.
Once the data is processed, observable features can be quantified