sslHiC

About

This repository contains the code and resources of the following paper:

Improving Comparative Analyses of Hi-C Data via Contrastive Self-Supervised Learning

Overviewer of the Framework

sslHiC is a computational framework for comparative analyses of Hi-C data, including reproducibility measurement and differential chromatin interaction (DCI) detection. sslHiC is pre-trained via contrastive self-supervised learning strategy. After pre-training, sslHiC projects genomic bins and chromatin interactions into latent feature embeddings, which subsequently can be used for the downstream comparative analyses.

Setup Environment

Setup the required environment using environment.yml with Anaconda. While in the project directory run:

conda env create

Activate the environment

conda activate sslHiC

How to Run

Reproducibility Measurement

python run_reproducibility_measurement.py --path1 ../example reproducibility/dataset/H1/ES_replicate1_hg19_50kb.cool --path2 ../example/reproducibility/dataset/H1/ES_replicate2_hg19_50kb.cool --resol 50kb --chr chr21

DCI Detection

python run_dci_detection.py --path1 ../example/DCI/dataset/GM12878_combined_chr21_2fold_c1.mtx --path2 ../example/DCI/dataset/GM12878_combined_chr21_2fold_c2.mtx --resol 50kb

--path1 and --path2 are the paths to the Hi-C data, supposing to be .cool/.mtx/.npy/.npz.

--resol is the resolution of the Hi-C data, supposing to be 500kb/50kb/10kb.

--chr is the target chromosome for analyses (required if using .cool files).

--key is the key of the data (required if using .npz files)

Running Examples

We provide running examples for our proposed sslHiC-Rep and sslHiC-DCI in the examples directory.

Data Simulation

Simulation of Noisy Contact Matrix

python simulate_noisy_contact_matrix.py --ref_path ../example/reproducibility/dataset/H1/ES_replicate1_hg19_50kb.cool --resol 50kb --chr chr21 --noise_type DropEdge --out_dir '.'

--ref_path is the path to the reference contact matrix, cooler file is required.

--resol is the resolution of the contact matrix.

--chr is the target chrosomoe for simualtion

--noise_type is the noise type used in simulation

--out_dir is the output directory for simulated contact matrices in cooler format.

Simulation of True DCIs

We employed the generateSimulation function from the FIND package (https://bitbucket.org/nadhir/find/src/master/) to simulate true DCIs. We followed the default setting of the generateSimulation function.

Baselines

The codes and guidelines of the baseline methods used in our work can be found at the following links:

GenomeDISCO: https://github.com/kundajelab/genomedisco.

HiCRep: https://github.com/TaoYang-dev/hicrep and https://github.com/cmdoret/hicreppy.

HiC-Spector: https://github.com/gersteinlab/HiC-spector.

Selfish: https://github.com/ay-lab/selfish

multiHiCcompare: https://github.com/dozmorovlab/multiHiCcompare.

License

sslHiC is licensed under the Apache License, Version 2.0: http://www.apache.org/licenses/LICENSE-2.0.