Capricorn

Dependencies

• Python 3.9
• Pytorch 2.0.1
• Numpy 1.24.3
• Scipy 1.10.1
• Pandas 1.5.3
• Scikit-learn 1.2.2
• Matplotlib 3.7.1
• tqdm 4.65.0
• Imagen-pytorch 1.25.11
• cooler 0.9.3
• einops 0.7.0
• einops-exts 0.0.4
• hic-straw 1.3.1
• chromosight 1.6.3

Data preprocessing

1. Set the enrivonment variables

• In dataset_information.py, set your root directory for data. For example, let the root directory be /data/hic_data.
• 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.
  • multichannel_matrix_dir: stores the transformed hic matrices with additional views
  • data_dir: stores the data for training, validation and test.

1. Download raw HiC data.

• Download the raw HiC data from GSE62525 GEO accession(Rao et al. 2014). We used GM12878 primary intrachromosomal and K562 intrachromasomal.
• Create the RAW_dir directory under your root directory. Unzip the raw HiC data into the directory. This would create a new directory with the cellline name under the directory.

1. Run the all-in-one scrips to generate data for training, validation and testing

• The scripts will run several scripts to get the data for training, validation and test:
  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 downsample them.
  3. Transform both high-coverage and low-coverage matrices to include additional views.
  4. Split the transformed matrices to get 40 by 40 patches, which can then be used for training, validation and test.
• See the codes for more details.

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

Run Capricorn

1. Calculate channel weights

• We have prepared the channel weights in ./weights.
• For the explanation of each weight file and how to get them, see the README in ./weights.

1. Train the model

• This will train our model with the default parameters on GM12878. The checkpoints and train details are saved under ./checkpoints/{start_time}_Capricorn_GM12878. You could replace GM12878 with the cellline you need.

1. Enhance HiC matrices with the model

• This will use our trained model to enhance the HiC matrices on another cell-line. The results are saved in the same directory as the checkpoint.
• Replace checkpoint_name, data_file and CELLLINE according to your need.

Evaluating results

1. Test the metrics

• We provide a all-in-one evaluation scripts. See HiC_evaluation/batch_evaluation.py for more details.
• In the scripts, there are evaluationg settings we used for the experiments in the Paper. You should modify the settings to meet your need.
• The following command will call the Example experiment, which evaluating the low-coverage matrices(resolution 10kb_d16_seed0) with loop F1 scores on GM12878 cell line.