HiCMamba: Enhancing Hi-C Resolution and Identifying 3D Genome Structures with State Space Modeling

1. Download raw Hi-C data and set the environment variables

• Download the raw Hi-C data from GSE62525. GM12878 and K562 are used in this work.
• Create the RAW_dir directory under your root directory and then unzip the raw Hi-C data into this directory.
• Set the related variables in dataset_information.py, including root_dir and RAW_dir.
• Also, set the directory name to store different data file.
  • RAW_dir: stores raw hic data
  • hic_matrix_dir: stores the hic matrices in npz format.
  • data_dir: stores the data for training, validation and test.

2. Run data_processing.Preprocess.py to process the raw Hi-C data and generate data for training and testing

• The script consists of four key step:

  1. Read raw data files from RAW_dir, save them in numpy matrix style(.npz files) in hic_matrix_dir.
  2. Read the high-coverage numpy matrices and downsampling them.
  3. Data normalization.
  4. Data division, transforming the data into 40 * 40 submatrices.

• The script can be executed using the following command:

python -m data_processing.Preprocess -c GM12878
python -m data_processing.Preprocess -c K562

The well preprocessed data is accessible at Google Drive. You can use the preprocessed data through downloading the data and then moving this data into data_processing directory.

1. Training

python train.py

2. Testing

python test.py

• Python 3.8.18
• Pytorch 1.13.0+cu117
• causal-conv1d 1.0.0
• mamba_ssm 1.0.1
• Numpy 1.24.4
• Scipy 1.10.1
• Pandas 2.0.3
• Scikit-learn 1.3.2
• Matplotlib 3.7.5
• tqdm 4.66.2

The causal-conv1d and mamba_ssm are strongly recommended downloaded from the BaiduNetdisk Link provided by VM-UNet. And then install the packages using:

pip install xxx.whl

We express our gratitude to the authors of HiCARN for sharing their open-source code.