Regional Synaptic Connectivity from Connectome Datasets

Alongside neuron-to-neuron connectivity (synapsed_to), each EM connectome carries regional connectivity: for every neuron, a summary of how many synapses it makes in each neuropil region — that is, the neuron’s input and output synapse tallies per region, rather than per partner neuron. This lets a neuron be characterised by where in the brain or VNC its inputs and outputs lie (for example “this cell has most of its postsynaptic sites in AL_R and its outputs in LH_R”), and lets regions be compared by how much synaptic traffic they carry.

What is loaded

For each neuron that VFB loads for a dataset — the cell-typed, non-deprecated neurons cross-referenced to that dataset’s Site — VFB records the neuron’s synapse counts in each region as edges from the neuron to the corresponding region individual (the same side-specific ROI individuals described in Neuropil Regions). Because each region individual is side-specific (part_of a body side), left/right is carried by the target region and does not need to be stored on the edge.

A neuron gets an edge for every region that resolves to a loaded region individual. ROIs with no loaded individual are dropped — e.g. the fine-grained optic-lobe columns (thousands of ME/LO/LOP columns that have no individual), FlyWire’s unassigned bucket (UNASGD), and neuprint’s NotPrimary pseudo-ROI — and are flagged in the build’s ROI-mapping report.

Counts are multi-level (nested), not a single partition

The region individuals VFB loads span several nesting levels — coarse grouping neuropils (e.g. SNP, INP, CX), the primary neuropils within them (e.g. SLP, SIP, AL), and finer subdivisions (antennal-lobe glomeruli, mushroom-body lobe slices, fan-shaped-body layers, ellipsoid-body domains). A region-connectivity edge is emitted at every level that has a loaded individual, and the counts are nested: a synapse is counted at its finest region and at each of its ancestors, so a parent region’s count equals the sum of its children (for example AL(R)’s upstream equals the sum over its glomeruli). This mirrors how the source data (neuprint’s roiInfo) reports counts at every level of the ROI hierarchy.

This makes the tallies multi-resolution rather than a sum-safe partition: they are correct per region at whichever granularity you query, but summing across nested regions double-counts the shared synapses. Consumers that need a non-overlapping total should pick a single level (e.g. the primary neuropils) rather than adding a parent and its children together.

Representation

Each neuron–region edge is one of two relations, carrying the counts as edge properties:

Relation	Direction	Counts
has postsynaptic terminal in (`RO:0002110`)	inputs the neuron receives in the region	`upstream` — number of the neuron’s postsynaptic sites (inputs) in the region
has presynaptic terminals in (`RO:0002113`)	outputs the neuron makes in the region	`Tbars` — number of presynaptic release sites (T-bars); `downstream` — number of downstream synaptic connections

The counts are stored as numeric edge properties, matching the representation VFB already uses for these regional tallies. A presynaptic edge is only asserted where the neuron actually has output structure in the region (a T-bar count for the neuprint datasets, or an output-synapse count for FlyWire), and likewise a postsynaptic edge requires an input count — so an edge always means the neuron genuinely has terminals of that polarity in the region.

Datasets and sources

Dataset	VFB symbol	Count source	Regions
Hemibrain	hb	neuprint `hemibrain:v1.2.1` `roiInfo`	brain neuropils (incl. glomeruli, MB slices, FB layers, EB domains)
MANC	mv	neuprint `manc:v1.2.1` `roiInfo`	VNC neuropils, tracts and nerves
male-CNS	mc	neuprint `male-cns:v1.0` `roiInfo`	brain + VNC neuropils
Optic-lobe	ol	neuprint `optic-lobe:v1.0.1` `roiInfo`	optic-lobe neuropils and layers
FAFB (FlyWire)	fw	FlyWire Codex neuropil synapse table (materialization 783)	lateralised brain neuropils

The neuprint dataset version is taken from the VFB Site node for each dataset (the dataset= in its link_base), so the regional counts always come from the same release VFB is pinned to — see Dataset Versions and Deprecation.

Sources

neuprint (neuprint.janelia.org) — for the Janelia datasets (Hemibrain, MANC, male-CNS, Optic-lobe), the per-region input/output tallies come from each neuron’s roiInfo (upstream, pre/T-bars and downstream per ROI).
FlyWire Codex (codex.flywire.ai) — for FAFB, the counts come from the Codex per-neuron neuropil synapse table (input and output synapses per neuropil). The file is the public Google Cloud Storage object https://storage.googleapis.com/flywire-data/codex/data/fafb/783/neuropil_synapse_table.csv.gz (materialization 783). FlyWire does not annotate T-bars, so its presynaptic edges carry the output-synapse count as downstream (no Tbars); its postsynaptic edges carry the input-synapse count as upstream.

Relationship to neuron-to-neuron connectivity

Regional connectivity is a per-region summary, not a per-partner one: it says how many synapses a neuron has in a region, but not which other neurons it connects to there. For neuron-to-neuron connections (with partner identities and weights) see the synapsed_to connectivity described under Versions & Deprecation. The two are consistent — the regional tallies aggregate the same underlying synapses — and both are refreshed together when a dataset version changes, following the same deprecation rules (counts are only loaded for non-deprecated neurons).