We developed a novel and simple computational method based on deep learning to enhance the resolution of Hi-C data.
Recovering high-resolution Hi-C data from low-resolution Hi-C data
python (3.6)
Tensorflow (v.1.13.1)
We recommand use the [anaconda3/5.2.0] (https://www.continuum.io) distribution to install the Dependency.
In the data processing stage, we combined the corresponding high-resolution Hi-C sub-matrix and low-resolution Hi-C sub-matrix into a sub-matrix in which its shape is (X1,y1+y2),where x1 is the abscissa of low-resolution and x2 is the abscissa of high-resolution, y1 and y2 is the ordinate of low-resolution sub-matrix and high-resolution sub-matrix,respectively. For example, if the low-resolution sub-matrix is (40,40) and high-resolution sub-matrix is (28,28), so the combined sub-matrix is (40,68).
In the training stage, the input matrix shape should be in the shape as (N, n1, n2,1), where N is the number of the input matrix, n1 and is the size of the combined sub-matrix and 1 is the number of channel.
In the specific code running process, please manually modify the corresponding parameters of training input file directory path,valid file and model save directory path in the SRHiC_main.py script under the src file. Last but not least,please set the training parameter in the main function to True. If you have modified the above parameters, then you can run directly
python SRHiC_main.py
Just low-resolution Hi-C samples are needed. The shape of the samples should be the same with the training stage. The prediction generates the enhanced Hi-C data, which is a bunch of sub-matrices. The user need to recombine them into a big matrix.
If you want to use your trained model or a model directly from the model folder in my repository, please manually modify the corresponding parameters of test input file directory path, model save directory path and the path of model checkpoint file in the SRHiC_main.py script under the src file. Last but not least,please set the training parameter in the main function to False. If you have modified the above parameters, then you can run directly
python SRHiC_main.py