FlyScope · Connectome Explorer
Inside a neuron, where inputs land reflects connections elsewhere in the network.
02 / The question
A connectivity graph draws each neuron as a point. Inside, every input lands somewhere on the cell’s branches. Does that internal arrangement record connections elsewhere in the network?
Fix a connection. Partner B contacts receiving neuron C: five or more records one way, fewer than five back.
Map the inputs. Each source input on C is measured along C’s own branches to the nearest input from B in the same brain region.
Read the far edge. Sources wired to B, an edge that never touches C, land nearer B’s inputs. Input maps alone then predict A→B for unseen sources.
Held fixed receiver and brain region · paths along C’s own branches · source cells withheld from fitting
Schematic04 / Paper findings
From network connections to input locations.
Open paper exampleSource APartner B
Move from a circuit to its receiving neuron.
Schematic05 / Results
The same relationship in MaleCNS and FlyWire, under three increasingly strict tests.
01 / Between sources
Mean path from a source input on C to the nearest partner input: sources that also contact B against all other sources.
Negative in 82.7% of 1,079 MaleCNS and 80.7% of 720 FlyWire receivers; 13,318 and 5,807 connections.
02 / Fixed inputs
Source A, receiver C and A’s inputs held fixed; partners that receive A input against partners with none, at three reference sites each.
Same ordering in 749 of 922 MaleCNS and 483 of 589 FlyWire receivers.
03 / Unseen sources
Predicting A→B for 200 test sources disjoint from 200 training sources, from input maps on known receivers.
Log loss 0.184 → 0.165 and 0.142 → 0.125; 30,702 and 21,822 scored pairs. Point estimates from one source split per specimen.
Beyond the fly brains, two adult C. elegans reconstructions also gave negative within-source differences (10 and 13 receivers), and a Platynereis larval population was too sparse to decide. These are associations within single specimens: anatomical opportunity, selective wiring, or both remain compatible.
Find a neuron. Follow its connections.
Interface preview · schematic arbor
Choose a dataset. Explore its connections.
08 / The paper
A neuronal connectivity graph records which cells are connected but discards the positions at which their inputs enter a receiving neuron. We asked whether this internal arrangement contains information about connections between the input neurons themselves. In two adult Drosophila connectomes, inputs from sources connected to a descending-neuron partner lay nearer that partner’s inputs on a shared receiver. Mean path-distance differences were 9.4 µm in MaleCNS and 8.0 µm in FlyWire. With source-input positions fixed and reference-site counts standardized, the relationship persisted across alternative partners. We then used source-labelled input maps on known receivers to predict connections between source cells excluded from model fitting. Adding spatial-distribution and distance features to input counts and receiver size increased average precision from 0.089 to 0.161 in MaleCNS and from 0.069 to 0.114 in FlyWire, with lower probability loss in both datasets. Predictions averaged information across eligible shared receivers. These findings connect the organization of inputs within neurons to connectivity among their sources and show that this spatial information is useful for inferring connections between unseen input cells.