This is the supplementary data and code for the paper:

Panser K, Tirian L, Schulze F, Villalba S, Jefferis GSXE, Bühler K, Straw AD. (2016) Automatic segmentation of Drosophila neural compartments using GAL4 expression data reveals novel visual pathways. Current Biology.

Please check our website at https://strawlab.org/braincode for updates and an interactive data browser. Updated versions of this source code may be found at https://github.com/strawlab/braincode.

This directory contains three subdirectories:

• braincode - the source code
• clustering-data - the results of the published clusterings and agglomerations
• 3d-models - models of the segmented VPNs, optic glomeruli and so on.

There are three types of data files included in the clustering-data directory:

• Samples HDF5 file (data/<dataset>/<region>/<region>_samples.h5)
• Clustering result volume file (data/<dataset>/<region>/<clustering_type>/<region>_clusterimage.nrrd)
• Cross-reference file (data/<dataset>/id_driver_image.csv)

These .h5 source files are in the HDF5 format. They contain the expression data used for the clustering. They have been subsampled from the original registered confocal stacks and have been taken from the brain region on which the clustering was performed.

Internally, the structure of each .h5 file is

/
├── ids
│   ├── 247_217_115
│   ├── 247_217_118
│   └── <x>_<y>_<z>
├── positionKeys
├── size
├── stepXY
└── stepZ

The ids group contains many individual datasets named <x>_<y>_<z>, where <x>, <y>, and <z> are the coordinates of the sampled voxel. Each dataset is a list of ids whereby each id corresponds to a particular confocal stack and driver line. The criterion for an id to be listed is that the expression in that stack must exceed a threshold value in that voxel.

The positionKeys dataset is a comma separated string of all considered positions within the given region. The coordinates are not in the downsampled space.

The size dataset species the number of voxels in the original (not-downsampled) volume.

The stepXY and stepZ datasets specify the amount by which the resulting clustering result volume (.nrrd file) has been downsampled.

These .nrrd files contain the volumetric coordinates of each cluster identified by the clustering algorithm in the NRRD format. Each voxel is assigned zero (for no cluster) or an integer that defintes the cluster number to which the voxel belongs.

This file contains the identity between an integer id, the file name of a confocal stack corresponding to that id, and the driver line from which the confocal stack was made.

• kmedoids_salspaugh.py The kmedoids algorithm
• dice.py The dice coefficient algorithm
• calculate_distance.py Computes the voxel-to-voxel distance matrix for a given brain region
• perform_clustering.py Run the kmedoids clustering algorithm
• util.py Various utilities
• plot_distance_matrix.py Create a plot showing the voxel-to-voxel distance matrix for a given clustering result
• fragments_per_cluster_step1_compute.py Compute which driver lines are expressed in which clusters
• fragments_per_cluster_step2_save_csv.py Save CSV file with the data per driver line
• fragments_per_cluster_step3_csv_to_json.py Convert CSV file with all rows to a JSON file with only significant rows
• calculate_cluster_stats.py Measure distances between medoids (inter-cluster) and between voxels in cluster (intra-cluster)
• save_cluster_info_json.py Save cluster statistics
• stability.py Evaluate the stability of clusterings