CDACHIE

Abstract

This repository contains the code for the paper "CDACHIE: Chromatin Domain Annotation by Integrating Chromatin Interaction and Epigenomic Data with Contrastive Learning."

CDACHIE (Chromatin Domain Annotation using Contrastive Learning for Hi-C and Epigenomic Data) is a method for identifying chromatin domains from Hi-C and epigenomic data. Our approach leverages contrastive learning to generate aligned representative vectors for both data types at each genomic bin. The concatenated vectors are then clustered using $K$-means to classify distinct chromatin domain types.

Installation

1. Clone the repository

git clone https://github.com/maruyama-lab-design/CDACHIE.git

2. Install the required packages in the environment

pip install -r requirements.txt

Data

We used data from the GM12878 cell line for our experiments.
The data directory structure is as follows:

data
├── input
│    ├── epigenomic_feature
│    │    ├── GM12878_100000_bins.txt
│    │    ├── *.bigwig (downloaded via make_feature.ipynb)
│    │    └── signals_1kb.npy (also generated by make_feature.ipynb)
│    └── hic_embedding
│         └── hic_line_embedding_128.csv
└── output
     ├── CDACHIE.bed
     └── clusters.csv

Epigenomic signal

Before executing the main script cdachie.py, bigwig files and their compiled file, signals_1kb.npy, supposed to be stored in the directory, data/epigenomic_feature, should be prepared with make_feature.ipynb.

Hi-C embedding

The Hi-C embedding data file (hic_line_embedding_128.csv) was created using the code from: https://github.com/nedashokraneh/IChDA/blob/master/src/dataset_class.py

Contrastive Learning & Clustering

Only proceed with this step after completing the data preprocessing steps above.

1. Configure your parameters in config.yaml
2. Run the script:

python cdachie.py

3. The annotation results will be saved to data/output/clusters.csv

Usage

To use CDACHIE for your own data:

1. Prepare your Hi-C and epigenomic data in the required format
2. Modify the configuration in config.yaml as needed
3. Run cdachie.py
4. The output file contains chromatin domain annotation

Archival Code

An archival snapshot of the code used in this study is available on Zenodo: https://doi.org/10.5281/zenodo.15751780.