Cell Type Coverage Analysis for Split-GAL4 Lines

Scripts to evaluate manually annotated matching hits and extrapolate the cell type discovery curve to the complete set of split-GAL4 lines in the collection described in Meissner et al. 2025.

For more information, please see the corresponding paper: Neuron types and distribution of a Drosophila driver line collection.

Annotated EM body ID-to-line matches

data/all_body_id_annotations_*.csv contain manually annotated matches between EM body IDs and split-GAL4 lines, including confirmed matches and highly ranked false matches from the automated search. Note that the corresponding paper primarily focuses on cell type-to-line matches, whereas this resource covers individual EM body IDs. These annotations are intended as a resource to facilitate future method development for automated matching. However, the dataset is not yet ML-ready and benchmark-ready, as it lacks, for example, dedicated documentation (e.g., a datasheet), predefined training splits, and an evaluation protocol.

Installation

git clone https://github.com/Kainmueller-Lab/split_gal4_lines_coverage_analysis.git
cd split_gal4_lines_coverage_analysis
pip install -r requirements.txt

Configuration

The scripts read their settings from config.toml (pass a different file with --config; the default is config.toml in the current directory). It contains:

• [general]: log_level and your neuprint_token. Replace the placeholder with your own token for neuPrint before running analyze_coverage.py, and do not commit it.
• [data]: paths to the annotation files (data/annotated_cell_types_*.xlsx) and the region and cell type mapping files in data/.

Commands

Run the coverage analysis

Computes discovery curves for the brain or VNC (or a specific region) and saves them as pickle files.

python src/analyze_coverage.py \
    --cns_part brain \
    --config <path-to-config-file> \
    --output_folder <path-to-output-folder> \
    --skip_existing \
    --num_subsets 100 \
    --num_sample_obs 500 \
    --num_sample_opt 100 \
    --num_sample_ext 100

Argument	Description
--cns_part	brain or vnc
--config	Path to the config file
--output_folder	Where the pickle files are written
--skip_existing	Skip regions that already have results saved as pickle
--num_subsets	Number of randomly sampled subsets from all annotated lines, used to create the corresponding number of models
--num_sample_obs	Number of random orderings used for Monte Carlo sampling to create the empirical discovery curve of annotated lines
--num_sample_opt	Number of discovery curves simulated during optimization for alpha
--num_sample_ext	Number of discovery curves simulated per model for final extrapolation

Visualize the results

Creates plots per region from the saved pickle files.

Per-region plots (brain)

python src/visualize_coverage.py \
    --input_folder <path-to-brain-output> \
    --cov_excel <path-to-brain-output>/coverages_brain.xlsx \
    --output_folder <path-to-plot-output-folder> \
    --cns_part brain \
    --do regions

Observed discovery curves (brain and VNC)

python src/visualize_coverage.py \
    --input_folder_brain <path-to-brain-output> \
    --input_folder_vnc <path-to-vnc-output> \
    --output_folder <path-to-plot-output-folder> \
    --cns_part both \
    --do dc_obs_both

Argument	Description
--input_folder	Folder with the pickle files from step 3 (single CNS part)
--input_folder_brain	Same, for the brain (used with --cns_part both)
--input_folder_vnc	Same, for the VNC (used with --cns_part both)
--cov_excel	Excel file with coverage values per region
--output_folder	Where the plots are saved
--cns_part	brain, vnc, or both
--do	Plot type: regions (per-region plots), dc_obs_both (observed discovery curves for brain and VNC), num_ct_per_line_both(number of cell types per line for brain and VNC), ct_freq_both (cell type frequency for brain and vnc)