CoolFace
Datasetpublic

SLOP011/flywire-fafb-connectome

FlyWire FAFB v783 Connectome — GNN-ready package The complete proofread wiring diagram of an adult female Drosophila melanogaster brain — 139,255 neurons and their synaptic connections — repackaged as a ready-to-train graph dataset. Companion to the flywire-gnn Python package. This is a dataset packaging of two public, no-auth sources: File Contents Source connections.parquet (474 MB) 15,091,983 unique directed neuron→neuron pairs at ≥1 synapse: pre, post, syn_count… See the full description on the dataset page: https://huggingface.co/datasets/SLOP011/flywire-fafb-connectome.

sourceHugging Facecc-by-4.0updated 5d agoView on Hugging Face
1likes428downloads
Dataset Card

FlyWire FAFB v783 Connectome — GNN-ready package

The complete proofread wiring diagram of an adult female Drosophila melanogaster brain — 139,255 neurons and their synaptic connections — repackaged as a ready-to-train graph dataset. Companion to the `flywire-gnn` Python package.

This is a dataset packaging of two public, no-auth sources:

FileContentsSource
connections.parquet (474 MB)15,091,983 unique directed neuron→neuron pairs at ≥1 synapse: pre, post, syn_count, and 6 synapse-weighted neurotransmitter probabilities (gaba, ach, glut, oct, ser, da)FlyWire v783 release (Zenodo, CC-BY-4.0): doi:10.5281/zenodo.10676866
nodes.parquet (2.9 MB)139,255 proofread neurons: root_id + annotations flow, super_class, cell_class, cell_sub_class, cell_type, hemibrain_type, hemilineages, top_nt, top_nt_conf, side, nerve, vfb_id, fbbt_idSupplementary Data 5 of Schlegel et al. 2024: doi:10.1038/s41586-024-07686-5
meta.jsondataset statistics, feature schema, task definitionthis package

Benchmark task: node classification on super_class

9 coarse classes (optic / central / sensory / visualprojection / ascending / descending / visualcentrifugal / motor / endocrine). Every neuron is labeled. The default benchmark graph uses directed edges at ≥5 synapses (the published FAFB "connection" threshold) → 2,700,513 unique directed edges.

  • —Node features (174-dim, wiring-derived): log1p(out_degree), log1p(in_degree), log1p(out/in synapses), log1p of out/in synapse counts across 79 neuropils, synapse-weighted mean NT profiles (in & out, 6 each).
  • —Edge attributes: log1p(syn_count) + 6 weighted NT probabilities.
  • —Splits: 70/15/15 train/val/test, stratified by label, seed 42.

Baseline results (real runs, CPU, seed 42)

ModelTest accuracyMacro-F1Best val acc (epoch)EpochsWall time
MLP (feat. only)0.98510.71150.9860 (200)200212 s
GCN0.91660.47020.9202 (180)2001414 s
GraphSAGE0.98120.75630.9820 (150)150783 s

What the benchmark shows: wiring-profile features alone nearly saturate accuracy (98.5%); message passing with mean aggregation (GraphSAGE) matches features and adds robustness on the rare classes (best macro-F1); GCN's normalized averaging over-smooths and trails. Report your model's accuracy and macro-F1 to compare on the same splits.

License & citation

Connectivity data is CC-BY-4.0. If you use this dataset, cite:

  1. 1.Dorkenwald et al., Neuronal wiring diagram of an adult brain, Nature 2024. doi:10.1038/s41586-024-07558-y
  2. 2.Schlegel et al., Whole-brain annotation and multi-connectome cell typing of Drosophila, Nature 2024. doi:10.1038/s41586-024-07686-5

Annotations redistributed here are from the Schlegel et al. 2024 supplementary data; the connectivity graph is derived from the FlyWire v783 public release. This package is not affiliated with the FlyWire Consortium. Explore interactively at codex.flywire.ai.