Tensor-Based Analysis of High-Order Chromatin Interactions in Pore-C Data

Initial setup:

• Use the following commands to install a Conda/Miniconda env for running TensorFox
• https://github.com/felipebottega/Tensor-Fox for more details

pip install TensorFox
conda create --name tfx_env --channel defaults jupyter numpy pandas scipy scikit-learn matplotlib numba IPython sparse_dot_mkl  

Protocol for creating and factorizing high-dimensional tensors

The corresponding Python scripts are in 'scripts/python_code' \

To create tensors for Pore-C data:

sbatch make_tensors.slurm

• You can check the log file make_tensors_log_slurm_{job_number}.txt for additional info
• You can change the parameters at the top of the make_tensors.py script
• You may have to modify the memory allocation in the job submission slurm script
  • Using a smaller window size will require more memory
  • Creating tensors for larger chromosomes will require more memory
  • Using a higher maximum cardinality will require way more memory

Make sure that make_tensors.slurm finishes before running the CPD job.
To perform CPD on the tensors:

sbatch cpd.slurm

• You can check the log file here: 'cpd_log_slurm_{job_number}.txt

To plot the factors, you can use the jupyter notebook 'scripts/jupyter_notebooks/cpd_results.ipynb/cpd_results.ipynb'

Protocol for creating and decomposing 2D tensors (i.e. Pore-C matrices with max cardinality = 2)

First, you will need to create a new conda environment with PyTorch, Cooler, CoolBox, and CoolTools.
For more information, refer to:

• https://gangcaolab.github.io/CoolBox/installation.html
• https://cooler.readthedocs.io/en/latest/quickstart.html#installation
• https://cooltools.readthedocs.io/en/latest/#

To create your conda env, use the following commands:

conda create -n cool_env python=3.10.12
conda activate cool_env
conda install -c conda-forge -c bioconda cooler cooltools bioframe coolbox numpy pandas h5py scipy
conda install pytorch=2.1.1 torchvision=0.15.2 torchaudio=2.1.1 cudatoolkit=11.3.1 -c pytorch

To create the 2D Pore-C tensors and create .cool files from the tensors:

cd /commons/groups/gursoy_lab/wlounsberyscaife/PoreC_Tensors/scripts/slurm_jobs
sbatch make_tensors_2D.slurm
# wait till make_tensors_2D is done running
sbatch 2D_tensors_to_cools.slurm

Note that the 2D_tensors_to_cools script will create a balanced .cool file and and unbalanced .cool file for each chromosome \

Then, you can play with the Jupyter Notebooks to create Pore-C matrices, decompose them, and annotate them with histone/CTCF data.

cd /commons/groups/gursoy_lab/wlounsberyscaife/PoreC_Tensors/scripts/jupyter_notebooks

• coolbox_plots.ipynb: uses CoolBox to create Pore-C matrices
  • Includes tracks for histone modification and CTCF enrichment
  • You will have to change the subrange depending on which chromosome you are plotting
• cooltools_plots.ipynb: uses CoolTools to perform Eigendecomposition on Pore-C matrices
  • You will have to modify the code depending on the chromosome you are plotting