SCW has 5 versions aimed at different situations.
We released two folders. One for one-step, the other for two-step.
In one-step, there are two subfolders. One for GAE, one for 2D-Guassian.
In two-step folder, there are three subfolders. One for GAE, one for 2D-Gaussian， for low resolution, and the other is 2D-Gaussian-high-resolution (recommended for resolutions larger than 100 Kbp, like 50 Kbp, or 20 Kbp).
- One-step 2D-Gaussian version is the simplest one. For running this, we need one Hi-C contact pair file, one chromosome length file.
One contact pair file is in the example_data file, which is GM12878 cell 12. The format for Hi-C is :
"1_0 766645 1_0 16241940";
Where "1_0" means the paternal of chromosome 1, "1_1" means the maternal of chromosome 1.
The format for the length file is
"1_1 249250621
1_0 249350632
2_1 ...
2_0 ... ";
Where "1_1" is the name of the chromosome, here is maternal chromosome 1, and "249250621" is the reference length that matched your data.
For runt the tool:
"g++ -o so2d -pthread -std=c++11 one_step_2D_Gaussian.cpp";
Where "so2d" is the compiled file, you can name it anything you like. Then run:
"./so2d -i ../../example_data/gm12.hic -res 1 -len ../../example_data/human_list -np 40 -o out_g12_1mb";
Where "-i " followed by the Hi-C file, "-res " followed by the resolution, the unit is Mbp, double value used here (0.5, 0.05, or 1, is the right style), "-np " followed by the thread numbers you will assigned, value type is integer, "-o " followed by the output file name.
2. One-step GAE version is requested for one more parameter, which is the imputation file.
We need to go to "GAE_package" in the main folder to generate that impuatation file. To generate an imputation file, we need to use GAE to generate the imputation matrix, then convert the matrix file for SCW. Go to "GAE_package" folder in the main folder, run:
"perl generate_edge.pl human_list 1 ../example_data/gm12.hic >gm12_edge";
to generate the node edge file first for GAE input, where "1" here represents the 1 Mbp resolution, it can be changed to any value.
Then run GAE for generating the imputation matrix by using:
"python gae.py --nodes 6106 --fin gm12_edge --fout gm12_mat --lr 0.001";
Where "--nodes" is the number of nodes, "--fin" is the edge file, and "--fout" is the output file for storing the reconstructed matrix. "--lr" is to set the learning rate. The more details you can see in the "readme" file in the "GAE_package" folder.
Here, we can convert the GAE imputed matrix to the SCW input we need by using:
"perl mat_to_hic.pl human_list 1 gm12_mat >gm12_imp.hic";
Where "1" is the resolution, "human_list" is the chromosome length file, and "gm12_mat" is the GAE output matrix.
Similar to one-step 2D-Gaussian, we run the following to generate the modeling:
"g++ -o sog -pthread -std=c++11 one_step_GAE.cpp";
"./sog -i ../../example_data/gm12.hic -res 1 -len ../../example_data/human_list -np 40 -imp ../../GAE_package/gm12_imp.hic -o out_g12_1mb";
Where "-imp" is followed by the GAE-imputed Hi-C file, the rest are the same as one-step-2D-Gaussian.
3. Two-step 2D Gaussian is the same as the one-step version.
"g++ -o st2d -pthread -std=c++11 two_step_2D_Gaussian.cpp";
"./st2d -i ../../example_data/gm12.hic -res 1 -len ../../example_data/human_list -np 40 -o out_g12_1mb";
4. Two-step 2D Gaussian high resolution is also the same.
"g++ -o st2dh -pthread -std=c++11 scw_two_step_high_resolution.cpp";
"./st2dh -i ../../example_data/gm12.hic -res 0.05 -len ../../example_data/human_list -np 40 -o out_g12_50kb";
5. Two-step GAE is similar to the one-step GAE version, but needs two GAE-imputed Hi-C files, one at 10 Mb resolution, and the other is the target resolution.
"g++ -o stg -pthread -std=c++11 two_step_GAE.cpp";
"./stg -i ../../example_data/gm12.hic -res 1 -len ../../example_data/human_list -np 40 -imp0 ../../GAE_package/gm12_imp_10mb.hic -imp ../../GAE_package/gm12_imp_1mb.hic -o out_g12_1mb";
Where "-imp0" is followed by the 10 mb GAE imputed Hi-C, "-imp" is followed by the target "-res" resolution.
6. Visualization.
The output from our tool is 3D coordinates like:
"3565.88 2252.64 2601.2 1_1 1000000-2000000
3564.76 2250.88 2599.2 1_1 2000000-3000000
..."
The first three columns are X, Y, and Z coordinates, "1_1" is the maternal chromosome 1 (for example, "22_0" means the paternal chromosome 22), the last column is the beads position.
In the output file, the last 46 lines (this number depends on how many chromosomes in your hic file, could be 20, 23, 40, or 46) are the chromosome summary with the bead numbers at the target resolution, which are shown as :
"1_1 230
1_0 228
2_1 242
2_0 242
... ";
We provide the tool for converting the original 3D coordinates to pdb file. Before using it, you need to ignore the last chromosome order lines. We can simply use the following:
"head -n -46 out_g12_1mb > out_g12_temp";
Then go to the "example_data" folder, using:
"perl convert_single.pl perl convert_single.pl ../two-step-SCW/two-step-2D-Gaussian/out_g12_temp g12.pdb";
to generate the pdb file for Pymol (not included, you should download it yourself) to visualize.