Data

Source

flymsg uses the MaleCNS v1.0 connectome of an adult male Drosophila melanogaster, the first complete central nervous system (central brain, optic lobes and ventral nerve cord), produced by the Janelia FlyEM project team with Google Research, the MRC Laboratory of Molecular Biology and the University of Cambridge.

• Project page: https://janelia-flyem.github.io/male-cns/
• Imaging: focused ion beam scanning electron microscopy (FIB-SEM), 8 nm isotropic voxels
• Segmentation: Google flood-filling networks, then proofread at Janelia and Cambridge
• Release: v1.0, June 2026 (v0.9: October 2025)
• License: CC-BY 4.0. Any published use must cite the dataset (see Citation).

Files

flymsg fetch downloads three Apache Arrow Feather tables from gs://flyem-male-cns/v1.0/connectome-data/flat-connectome/ (served over HTTPS) into data/raw/. No neuPrint account or token is needed.

minconf-0.5 means synapses detected with confidence ≥ 0.5, the threshold used by the release. Other tables on the download page (synapse points, 12.7 GB; skeletons; EM volumes) are not used.

Compaction (flymsg build)

build keeps bodies with status == "Traced" and edges whose two ends are both traced.

The dropped bodies are orphans, glia, fragments and unassigned segments. Most of the raw synapse total sits between those fragments: traced neurons receive 130.4M input synapses, and 95.1% of them come from other traced neurons. The compacted graph therefore keeps almost all of the input to the neurons it models.

Reproduce: flymsg build prints the kept counts; the 95.1% comes from summing connectome-weights rows whose body_post is traced, split by whether body_pre is traced.

Schema

Row position is the neuron index used everywhere (edges.pre/edges.post, sim arrays). Neurons are sorted by bodyId.

data/neurons.parquet

data/edges.parquet: pre, post (neuron indices, int32), weight (synapse count, int32).

Known data limits

• One animal: no measure of variability between individuals.
• Chemical synapses only: gap junctions are invisible to EM at this resolution, so electrical pathways (e.g. Giant Fiber → PSI) are missing or underweighted.
• Predicted transmitters: consensus_nt is a machine prediction. 502 traced neurons have none and 3,100 are unclear.
• Missing names: 2,605 traced neurons have no type and 7,040 no instance.
• Incomplete neurons: 3.2 % of bilateral types have one side far below the other (see Reconstruction completeness).

Reconstruction completeness (flymsg audit, runs/audit.txt)

Proofreading can leave a neuron incomplete, and a bilateral pair makes it visible: the two sides should carry similar synapse counts. flymsg audit compares, for each of the 11,095 cell types with neurons on both sides (side from the instance suffix), the median input synapses of the right and left neurons, and scores log2(R/L) against all types with a robust z-score (median/MAD). Across types the 5–95 % range of log2(R/L) is −0.47 … 0.53; 355 types (3.2 %) are flagged at |z| > 3. For the neurons behind flymsg's results it also reports the share of each neuron's input that comes from traced partners, and the same pair's asymmetry in FAFB where the type is matched.

The audit sees totals, not single connections: GF → TTMn has 70 synapses on the right and 20 on the left while both GF and TTMn totals are symmetric, an asymmetry it cannot attribute. Rule: a result that hinges on a flagged neuron must say so; where a bilateral pair is flagged, flymsg reads the side that is not.

Eye map (flymsg eye)

The viewing direction of each lens comes from Zhao et al. 2025 (doi:10.1038/s41586-025-09276-5), repository reiserlab/eyemap_T4 at commit 99d2a43: data/eyemap.RData (lens hex coordinates) and data/microCT/20240701.RData (lens directions). Both are GPL-3.0. flymsg eye downloads them into data/raw/eyemap/ on first use; flymsg does not redistribute them or anything derived from them. See eye.

Odor responses (flymsg odor)

Odorant responses come from the Database of Odorant Responses (DoOR; Münch & Galizia 2016, Sci Rep 6:21841, doi:10.1038/srep21841), repository ropensci/DoOR.data at db323a4: door_response_matrix.csv (consensus response per odorant and receptor, normalised to [0, 1]), door_mappings.csv (receptor → glomerulus) and odor.csv (names, InChIKeys). They are CC BY-SA 4.0. flymsg odor downloads them into data/raw/door/ on first use; flymsg does not redistribute them or anything derived from them.

Medulla voltages (flymsg.medulla)

Groschner, Malis, Zuidinga & Borst 2022, Nature 603:119, doi:10.1038/s41586-022-04428-3, Fig. 3a: whole-cell traces of Mi1, Tm3, Mi4, Mi9 and C3 to 30°/s ON and OFF edges, deposited on Edmond (doi:10.17617/3.8G, version 1.0, CC BY 4.0). The T4/T5 v3 run fetches the five files into data/raw/groschner2022/ and refuses one whose sha256 differs from the pinned value in src/flymsg/medulla.py; they are fetched, not redistributed.

MANC muscles (flymsg muscles)

The muscle of each motor neuron type comes from MANC: Cheong, Eichler, Stürner et al., "Organization of circuits linking descending input to motor output in the Drosophila Male Adult Nerve Cord connectome", eLife 13:RP96084 (2024), doi:10.7554/eLife.96084, CC BY 4.0. flymsg muscles downloads its Supplementary file 3 ("MN and efferent light-level matching in MANC", 834 rows) from https://cdn.elifesciences.org/articles/96084/elife-96084-supp3-v1.csv into data/raw/manc/ and refuses a file whose sha256 is not c46e3cec1a5114b8d981d5f0383012c8e7fcdebd6c4e3eaffe4ac3ba1def2e94. It writes data/motor/manc_types.parquet (one row per MANC type: targets, muscle, certainty range, publications, source) and data/motor/muscles.parquet (one row per MaleCNS vnc_motor neuron, joined on the mancType annotation of the raw body annotations). See embodiment for the match counts.

Female brain: FlyWire FAFB v783 (--dataset fafb)

flymsg --dataset fafb fetch and build write the female brain to data/fafb/ in the same schema, for the comparisons in comparison.md. Sources, pinned to commits:

Mapping: type = cell_type (also flywireType), instance = type + side, superclass = super_class, class = cell_class, subclass = cell_sub_class, entryNerve = nerve, fruDsx = fru_dsx, nt = top_nt. sign follows the same transmitter map as MaleCNS; shiu_sign keeps the published model's own signs (it treats dopamine, serotonin and octopamine as ±1 and used other transmitter predictions: the two differ for 14,515 neurons), so that the model can be reproduced exactly.

FAFB is brain only (no VNC) and one female. Shiu et al. used FlyWire v630; of their 21 sugar GRN IDs, 20 still exist in v783. The published model's side labels are mirrored with respect to the FlyWire side column: its "sugarR" GRNs are annotated left and its "MN9 left" is CB0701_R.

Citation

Berg, S. et al. Sexual dimorphism in the complete Drosophila male central nervous system connectome. Cell 189, 5504–5526 (2026). doi:10.1016/j.cell.2026.08.015. Preprint (different title: "Sexual dimorphism in the complete connectome of the Drosophila male central nervous system"): bioRxiv, doi:10.1101/2025.10.09.680999.

Take the full author list from the Cell article.

For --dataset fafb (FlyWire), as the sources ask, also cite:

• Dorkenwald, S. et al. Neuronal wiring diagram of an adult brain. Nature 634, 124–138 (2024). doi:10.1038/s41586-024-07558-y
• Schlegel, P. et al. Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature 634, 139–152 (2024). doi:10.1038/s41586-024-07686-5
• Matsliah, A. et al. Neuronal parts list and wiring diagram for a visual system. Nature 634, 166–180 (2024). doi:10.1038/s41586-024-07981-1
• Shiu, P. K. et al. (2024), for the model's connectivity and signs (see validation).
• Berg et al., the MaleCNS paper above, for the dimorphism and fru_dsx annotations.