ChromVAE is a deep learning method to analyze chromatin conformations obtained from single-cell imaging studies.
ChromVAE has two major contributions:
-
The latent space obtained from chromVAE monitors the progression of chromatin folding process.
-
The probability estimated from chromVAE provides chromatin energy landscape.
More details can be found in the paper Characterizing Chromatin Folding Coordinate and Landscape with Deep Learning by Wen Jun Xie, Yifeng Qi and Bin Zhang.
The package has been tested on CentOS Linux release 7.6 with the following software: Python 3.7, PyTorch 1.2, Numpy 1.16, Pickle 4.0
The chromatin imaging data were downloaded from Bintu et. al., Science, 2018, 362, eaau1783. The distance map was then binarized to contact map provided in the ./data/ directory. 90-kb resolution was used and there are 378 chromatin contacts for the studies region.
-
./data/HCT116/: contact matrixes for wild-type cell -
./data/HCT116_auxin/: contact matrixes for cohesin-depleted cell
-
./script/VAE_combine_train.py: code used to train chromVAE -
./script/VAE_combine_latent.py: code used to get the latent space after training chromVAE
We also include the code to analyze the latent space in the ./analysis/analysis_latent.ipynb.