wjychem / ggl

GenomeGeometricLearning

Implementation of Characterization of network hierarchy reflects cell-state specificity in genome organization Adapted PyTorch implementation of Poincaré Embeddings for Learning Hierarchical Representations

Installation

Simply clone this repository via

git clone https://github.com/wjychem/GGL.git
cd GGL
conda env create -f environment.yml
python setup.py build_ext --inplace
source activate poincare

HiC2graph

To fulfill network inference and graph analyses from Hi-C datasets, first

mkdir Data graph_statistics hic

Then run HiC2graph_statistics.m. This will derive the closeness centrality and average shortest path length of chromatin contacts networks (CCNs) as well as the inputs for betweenness centrality calculation and Poincaré Embeddings.

Embed Chromatin Contact Networks in Poincaré disk

cd train_script/
./sub-train-hic.sh

This will train the embeddings and save the output results in the hic directory with the same name as input Hi-C datasets. Scripts in train_script directory contain the hyperparameter setting to reproduce the results for Characterization of network hierarchy reflects cell-state specificity in genome organization.The loci ids and embedded coordinates will be saved in the output folder as keys.*.txt and pe.coors.*.txt, respectively, which are used for visualization.

Dependencies

• Python 3 with NumPy
• PyTorch
• Scikit-Learn