Diffusion embedding

Contents

This repository contains some of my diffusion embedding (DE) work. The docs/ directory contains Jupyter notebooks with diffusion maps tutorial and a simulation demonstrating that DE can successfuly recover ovarlapping gradients of connectivity. The folder also contains plot_surfaces.py function for surface plotting in Pyhton.

The main directory includes diffusion_embed.py file, which contains code for performing DE.

Requirements

The code has been written in Python 2.7. The DE code has been integrated into the class object. Following modules are neccessary to run all elements of the code:

• numpy
• scipy
• pySTATIS ( http://github.com/mfalkiewicz/pySTATIS )

Usage

from diffusion_embed import Diffusion_Embedding as de

emb = de(source_path = '/home/me/awesome_project/timeseries/',
                   file_template = '%s_timeseries.npy',
                   subjects = ['22529','23197','23269','23464','23734','23884','24691','24715','24757'],
                   output_path = '/home/me/awesome_project',
                   mwall = False)

emb.compute_embeddings()
emb.realign_embeddings()
emb.project_template_subjects()

Following options are available for the Diffusion_Embedding class:

source_path (mandatory) - the path where timeseries are stored

file_template (mandatory) - format of files to be read in

subjects (mandatory) - the changing part of file_template

output_path (mandatory) - where to store output files

diff_time (optional, default 0 ) - diffusion time

diff_alpha (optional, default 0.5) - diffusion operator

diff_ncomp (optional, default 10) - number of components to save

subjects_subset (optional, default None) - subset of subjects to use for template creation and back projection

output_suffix (optional, default 'embedding') - suffix to add to output filenames

ftype (optional, default 'npy_timeseries') - type of input data

surf (optional, default 'fsaverage4') - FreeSurfer template for the data, in calse mwall = True

mwall (optional, default False) - if True, medial wall vertices will be removed from analysis based on surf parameter

tp (optional, default None) - timepoints to extract from timeseries

affinity_metric (optional, default 'correlation') - affinity metric to use between timeseries

realign_method (optional, default 'STATIS') - method of aligning subjects into a common space (basis)