allen-voxel-network

Tools for working with Allen Institute for Brain Science voxel-scale connectivity data.

Requirements:

• numpy
• scipy
• h5py
• allensdk

To fit connectomes:

• spatial-network-regression

Generating a voxel linear model

1. Edit run_setup.py. This sets which structures will be included, the values of the regularization parameter, etc.
2. python create_visual_matrices.py. This will create a hierarchy of directories for model fitting with nested cross-validation.
3. Run the commands in model_fitting_cmds (located in the project directory) to perform the model fits.
4. Run python model_select_and_fit.py. In the inner cross-validation loop, evaluate the errors and perform model selection.
5. Run the commands in model_fitting_after_selection_cmds. This will fit the selected models.
6. Run python region_model_fits_and_voxel_errors.py. This will both evaluate the errors of the voxel models as well as fit regional models and compare their errors to the voxel models.

Visualizing voxel model

1. Run python voxel_model_visualizations.py. This performs fake injections into VISp, plotting the results. Also saves volumetric data & region labeled plot.
2. You can turn the virtual injection pictures into a movie easily: avconv -q 4 -r 7 -b 9600 -i int_virt_inj%d.png movie.mp4
3. You can visualize the volumetric data in VTK format (.vti files). Use Paraview.

Generating a regional model

First edit the following scripts to set the data and output directories, then run:

 python create_regional_matrices.py
 python run_new_regional_model.py

If you want to compare the output of this model to that from Oh et al. (2014), this can be accomplished with compare_new_old.py.