Computing GW Distances

To compute the Gromov-Wasserstein (GW) distance between intracellular distance matrices, users can employ the function cajal.run_gw.compute_gw_distance_matrix().

This section assumes that the user has already obtained the intracellular distance matrices for their cells. It is worth noting that the GW distance can be calculated using the same function regardless of how the intracellular distance matrices were computed and whether they represent the Euclidean or geodesic metric.

To use the function, the user should provide the path to an input *.csv database containing the intracellular distance matrices through the argument intracell_csv_loc. The output GW distance matrix will be saved in a new .csv file specified by the argument gw_dist_csv_loc.

    intracell_csv_loc = "/home/jovyan/CAJAL/CAJAL/data/swc_icd.csv",
    gw_dist_csv_loc = "/home/jovyan/CAJAL/CAJAL/data/gw_dists.csv",
    num_processes = 8)

By default, the coupling matrices which represent the best possible pariting between two cells are not retained. However, cajal.run_gw.compute_gw_distance_matrix() also provides functionality to save the coupling matrices. This is for instance required for the computation of average cellular shapes.

Numpy should automatically parallelize the computation across multiple cores. Users on Windows can check the process manager, while those on Unix-based systems can use the “top” command to verify that all cores are being utilized.


Note that storing the coupling matrices will generate a large amount of data, which scales quadratically with the number of input cells. For example, if there are 150 cells with 50 sampled points each, the resulting database size may be approximately 180MB. File IO may also become a bottleneck in the computation. Therefore, users should exercise caution when saving the coupling matrices, especially when working with a large number of cells.