Polymer-Penalty

Code and derived parameters for the manuscript: A multi-component power-law penalty corrects distance bias in single-cell co-accessibility and deep-learning chromatin interaction predictions.

This repository provides a framework to correct the systematic distance bias found in proxy data of 3D genome architecture.

---

Workflow

The method takes biased proxy data and a reference Hi-C dataset as input. It fits a multi-component power-law model to the Hi-C data to derive a penalty function.

---

Installation

1. Clone the repository

git clone [https://github.com/jlab-code/polymer-penalty.git](https://github.com/jlab-code/polymer-penalty.git)
cd polymer-penalty

2. Install dependencies

pip install -r requirements.txt

---

Usage

Workflow 1: Applying Pre-computed Global Consensus Penalties

Use our parameters derived for Soybean, Rice, and Maize.

1. Place your co-accessibility scores in the data/ directory.
2. Open the Jupyter Notebook: scripts/apply_correction.ipynb.
3. Select your target species:

# USER: Select species and model type
SPECIES = "Soybean"          
USE_GLOBAL_CONSENSUS = True  

4. Run all cells.

---

Workflow 2: Deriving Custom Penalties for New Species

Use our GMM-pipeline to generate a custom model for any species.

1. Place your Hi-C loops in .bedpe format in the data/ directory.
2. Open the Jupyter Notebook: scripts/get_penalty_function.ipynb.
3. Update the data path:

hic_path = "../data/your_new_species_HiC.bedpe"

4. Run all cells.

---

Citation

A multi-component power-law penalty corrects distance bias in single-cell co-accessibility and deep-learning chromatin interaction predictions. (2026).

---