ConnectomeMatch

Python implementation of preprocessing and significance testing with inexact graph matching accross a population.

This is a Python implementation built upon the techniques used in Osmanlıoğlu et al (2019).

It also allows for multi-threading of the more consuming proccesses (density matching and permutation testing).

Usage

First install the package.

pip install ConnectomeMatch

Then import

import ConnectomeMatch as cm

Preprocessing

Density Matching

Sometimes we are matching groups of graphs with different densities. For instance a sparse DTI-MRI connectome against a fully connected MEG connectome.

Given two group of N graphs (shape=N x Node x Node), we threshold the sample matrix to match the target matrix

sample_matrix = cm.match_density(sample_matrix, target_matrix, 0.2, 0.01,100)

Matching

One function executes the matching on the group level. It normalises per graph pair and allows for built in multithreading.

 diagonal_matches, binary_matrices, euclidean_distances = cm.match_graphs(sample_matrix, target_matrix, njobs=4, log=True) 

the function returns the matches along each graphs diagonal, binary matrixes for each participant and the euclidean distance between nodes of each graph.

Inference

A permutation function generates a null distribution of binary matches for the sample. This can then be compared to the average matching accuracy from binary matching matrices above

import numpy as np
null = cm.generate_null_dist(sample_matrix, target_matrix,perms=100, njobs=4) # get null distribution
montecarlo_thresh = np.percentile(null, 0.95, axis=2) # get threshold values for matching matrix
significance_mask = binary_matrices.mean(axis=0) > montecarlo_thresh # mask for all significant values