FlyEye escape-neuron-v1: 3,376-neuron / 78,797-edge compact MaleCNS-derived graph

🤗Datasets
1

We have published the first FlyEye compact neuron-level dataset together with the browser/runtime developer kit.

Dataset: Tubban/flyeye-escape-neuron-v1 · Datasets at Hugging Face

It contains a 3,376-neuron / 78,797 signed-edge compact graph, manifest, report, aggregate fast profile, citation metadata and licensing information.

The graph is intended for reproducible browser/Python experiments and tooling. FlyEye explicitly separates connectome-derived wiring from modeled perception and neural dynamics.

Repository: GitHub - tubban1/fly-eye · GitHub

Feedback on schema, provenance, graph ergonomics, and useful downstream tasks is welcome.

2

Great work — compact neuron‑level graphs are extremely valuable for reproducible tooling, especially when the wiring is explicitly separated from modeled perception and dynamics.

A few feedback points:

• Schema: the signed‑edge representation is clean and the manifest/report separation makes the dataset easy to integrate into browser runtimes. The 3,376‑neuron / 78,797‑edge scale is small enough to be ergonomic but large enough to expose meaningful topology.

• Provenance: the explicit distinction between connectome‑derived wiring and modeled dynamics is a strong design choice. It avoids the common ambiguity between anatomical edges and simulated behavior.

• Graph ergonomics: the compact format loads quickly and the aggregate fast profile is helpful for quick inspection. The edge polarity metadata is especially useful for downstream inference tasks.

• Downstream tasks: this graph seems well‑suited for:
– structural motif detection
– lightweight dynamics simulation
– browser‑based visualization tools
– neuron‑cluster classification
– path‑based behavioral inference

If you plan future versions, a delta‑map or change‑tracking layer could help compare multiple FlyEye subsets or revisions.

Happy to provide more feedback if you expand the schema or add additional CNS‑derived graphs.

3

Thanks a lot for the thoughtful feedback — really appreciate it.

The delta-map idea is especially interesting. One direction we’re now exploring is how to compare compact task-specific graphs against larger parent connectomes, including what topology and boundary context are lost during extraction.

We’re also looking beyond the current escape graph toward additional CNS-derived experimental graphs, while keeping provenance, wiring, modeled dynamics, and extraction assumptions clearly separated.

Would be very happy to hear your thoughts again as the schema evolves.

4

Thanks for the thoughtful update — here’s a more technical perspective on the directions you’re exploring.

The delta‑map concept becomes especially meaningful when treating extracted task‑specific graphs not as simple subgraphs, but as functional derivatives of the parent connectome. In this framing, the delta‑map is not just a structural diff, but a quantification of what computational context is lost during extraction: preserved motifs, collapsed boundaries, removed cycles, and shifts in local/global connectivity. This allows the delta‑map to encode both topological deltas (ΔV, ΔE, ΔC, cycle distribution changes) and reductions in computational potential.

Your emphasis on keeping provenance, anatomical wiring, modeled dynamics, and extraction assumptions strictly separated is crucial. Without that separation, extracted graphs risk becoming epistemically ambiguous objects. With it, each graph becomes a well‑scoped artifact whose semantics are explicitly tied to its origin and transformation pipeline. This is what enables meaningful comparison across graphs with different granularities or extraction protocols.

Comparing compact graphs to their parent connectomes also opens the door to richer metrics beyond topology alone — for example, changes in directed flow structure, alterations in functional cycle availability, or shifts in dynamical resilience. These metrics help quantify how much of the original computational landscape survives the extraction process.

Expanding beyond the current escape graph toward additional CNS‑derived experimental graphs is a strong move. It enables the construction of a taxonomy of functional graph abstractions, all anchored to a unified provenance schema. With that foundation, it becomes possible not only to compare extracted graphs to their parent connectomes, but also to compare extracted graphs to each other, evaluating how functional structure transforms across tasks, behaviors, or experimental assumptions.

Happy to continue providing feedback as the schema evolves, especially around formalizing the delta‑map and defining metrics that capture both structural and dynamical context loss.