In this package, we provide four algorithms to infer the 3D structure of the genome: MDS, NMDS, PM1, PM2. For all methods, chromosomes are represented as beads on a string, at a given resolution. All are installed as a standalone program: pastis-mds, pastis-nmds, pastis-pm1, pastis-pm2.

Running the example

We provide a sample interaction count matrix of the first 5 chromosomes and configuration file config.ini. To run the example, edit the two options binary_mds (respectively binary_pm) to the path where you have installed MDS_all (respectively PM_all).

binary_mds: /bioinfo/users/nvaroqua/.local/bin/MDS_all
binary_pm: /bioinfo/users/nvaroqua/.local/bin/PM_all
resolution: 10000
output_name: structure.pdb
chromosomes: 1,2,3,4,5
organism_structure: files/budding_yeast_structure
counts: data/counts.npy

To run the code, simply call the program of your choice, and the repository containing the configuration file as argument. From the root of the repository, to run the MDS:

pastis-mds example

A bunch of files, necessary for the optimization are written in the same folder as the optimization, including the results of the optimization: mds.structure.pdb.txt and mds.structure.pdb. The txt file contains a the flatten array of coordinates (x_1, y_1, z_1, x_2, y_2, \dots) while the pdb contains a smoothed interpolation of the structure for visualization purposes.

Running the algorithms on your own structure

Running the algorithms on your own structure, and your own organism will require a bit more work.

The interaction count matrix needs to be saved as a numpy array, in the folder of your choice. Numpy arrays are memory-efficient container that provides fast numerical operations: the ideal structure to work efficiently with matrices in Python. For efficient and accurate

The organism structure (chromosomes lengths of the organism) have to be specified in a text file, one chromosomes length per base pair. See example/files/budding_yeast_structure for an example.

Then, create the configuration file, specifying the resolution on which to run the algorithm: the resolution of the interaction counts matrix, the chromosomes lengths and the resolution should be coherent.