flyconn

Research-grade Python toolkit over the public Drosophila connectomes (MaleCNS v1.0, FlyWire v630/v783, hemibrain, MANC, BANC): a harmonized offline data layer, signed sparse graphs, uncertainty propagation, a validated cross-platform LIF simulator, declarative in-silico experiments with reports, and cross-dataset comparison.

Status: pre-alpha; milestones M0–M7 of docs/PLAN.md are implemented and validated against published results on real data (docs/GOLDEN_RESULTS.md, docs/PROGRESS.md).

Install

pip install "flyconn[sim]"            # data, graph, uncertainty, simulator
pip install "flyconn[sim,interpret]"  # + connectome_interpreter adapter

The interpret extra resolves connectome-interpreter 2.9.5 from PyPI; flyconn's parity tests run against a newer git commit of that package (see ADR-0002), so treat the PyPI combination as untested until upstream releases again. Documentation: https://moiz-lakkadkutta.github.io/flyconn/

Quickstart

uv sync --extra sim --extra interpret --extra access
uv run flyconn data list
uv run flyconn data pull malecns@1.0 --level weights      # ~1.1 GB, checksummed, resumable
uv run flyconn data pull shiu@630 --level weights          # Shiu et al. 2024 model inputs (90 MB)
uv run flyconn run examples/specs/w2_malecns_lb3_silence_gng232.yaml --out runs/w2

Python:

from flyconn.data.store import Store
from flyconn.graph import ConnectivityMatrix, find_paths
from flyconn.uncertainty import path_stability

m = ConnectivityMatrix.from_store(Store.open("malecns@1.0"), min_weight=5)
lb3 = m.meta.index[m.meta["cell_type"].str.match(r"^LB3", na=False)]
mn9 = m.meta.index[m.meta["cell_type"] == "MN9"]
paths = find_paths(m, lb3, mn9, max_hops=3, min_edge_fraction=0.01, label="cell_type")
stab = path_stability(m, lb3, mn9, thresholds=[5, 10, 20], n_samples=100, seed=0)

Anchor workflows: examples/w1_pathways.py, examples/w4_version_drift.py, examples/specs/w2_*.yaml, examples/w3_male_vs_female.py, examples/w5_driver_lines.py. Scientific caveats: docs/caveats.md.

Not to be confused with the Cambridge FlyConnectome group's tools (cocoa, flywire_annotations); flyconn is independent and builds on that ecosystem.

Scientific stance

A connectome is wiring. Neurotransmitter identities are predicted, weights are synapse counts, and every simulation output is a model prediction. Controls run by default, uncertainty is propagated, and every result carries provenance and the citations of the datasets it used. See docs/caveats.md once written.

Licence and attribution

Code: Apache-2.0. Data: CC BY 4.0 by the respective consortia; every result object emits the citation list it depends on. Design ideas borrowed from MIT-licensed projects are listed in ATTRIBUTION.md.