Neuronal wiring diagram of an adult brain

To assess quality, a group of expert centralized proofreaders conducted a review of 3,106 segments in the central brain. These specific neurons were chosen based on certain criteria such as significant change since being marked complete and small overall volume. An additional 826 random neurons were included in the review pool as well. Proofreaders were unaware which neurons were added for quality measurement and which ones because they were flagged by a metric. We compared the 826 neurons before and after the review and found that the initial reconstruction scored an average F1-Score of 99.2% by volume (Extended Data Fig. 2a,b). F1-Score is defined as the harmonic mean of recall (R) and precision (P) with precision defined as the ratio of true positives (TP) among positively classified elements (TP + FP (false positives)) and recall defined as the ratio of TPs among all true elements (TP + FN (false negatives)).

Quantification of proofreading effort

Any quantification of the total proofreading time that was required to create the FlyWire resource is a rough estimate because of the distributed nature of the community, the interlacing of analysis and proofreading and the variability in how proofreading was performed. The second public release, version 783, required 3,013,513 edits. We measured proofreading times during early proofreading rounds that included proofreading of whole cells in the central brain. We collected timings and number of edits for 29,135 independent proofreading tasks after removing outliers with more than 500 edits. From these data, we were able to calculate an average time per edit. However, we observed that proofreading times per edit were much higher for proofreading tasks that required few edits (<5). That meant that our measurements were not representative for the second round of proofreading which went over all segments with >100 synapses. These usually required 1–5 edits. We adjusted for that by computing estimates for proofreading speeds of both rounds by limiting the calculations to a subset of the timed tasks: (round 1) The average time per edit in our proofreading time dataset, (round 2) the average time of tasks with 1–5 edits. We averaged these times for an overall proofreading time because the number of tasks in each category were similar. The result was an average time of 79 s per edit which adds up to an estimate of 33.1 person-years assuming a 2,000 h work year.

Attachment rates

We adopted the attachment rate (also referred to as ‘completion rate’) calculations from the hemibrain. Every presynaptic and postsynaptic location was assigned to a segment. Using the neuropil assignments, we then calculated the fraction of presynapses that were assigned to segments marked as proofread for each neuropil and analogous for postsynaptic locations.

Comparison with the hemibrain

We retrieved the latest completion rates and synapse numbers for the hemibrain from neuprint (v1.2.1). In some cases, neuropil comparisons were not directly possible because of redefined regions in the hemibrain dataset. We excluded these regions from the comparison.

Crowdsourced annotation

The large FlyWire community and diversity of expertise enabled us to crowdsource the identification of neurons. There is no limit to the number of annotations a neuron can receive. A standardized format is encouraged but not required. One user might first report that a neuron is a descending interneuron, whereas another might add that it is the giant fibre descending neuron, and yet another might add all its synonyms and citations from the literature. Contributors’ names are visible so they can be consulted if there is disagreement. The disadvantage to this approach is that there is no single precise name for every neuron, but the advantage is a richness of information and dialogue. The annotations are not meant to be a finished, static list, but rather a continually growing, living data source. These annotations were solicited from the FlyWire community through town halls, email announcements, interest groups in the FlyWire forum, online instructions, and by personal contact from the community manager. Citizen scientists also contributed annotations, after receiving training on particular cell types by experts.

Neuron categorizations

Neuron categorization, sensory modality annotations and nerve assignments are described in detail in our companion paper. In brief, neurons were assigned to one of three ‘flow’ classes: afferent (to the brain), intrinsic (within the brain) and efferent (out of the brain). Intrinsic neurons had their entire arbor within the FlyWire brain dataset. This included cells that projected to and from the SEZ. Next, each flow class was divided into superclasses in the following way. afferent: sensory, ascending. intrinsic: central, optic, visual projection (from the optic lobes to the central brain), visual centrifugal (from the central brain to the optic lobes). efferent: endocrine, descending, motor.

Quantification of intrinsic neurons

We define whether a neuron is ‘intrinsic’ to a region on the basis of its synapse locations, rather than its arbor. In other words, the neurites of an intrinsic neuron are allowed to exit the region, provided that they do not make synapses after leaving. Information about C. elegans synapse locations was obtained from the diagrams in White et al..

The ‘brain’ of C. elegans can be defined as the neuropil extending from the ring-shaped structure around the pharynx to the excretory pore. (We follow the authors who call this region the nerve ring plus the anterior portion of the ventral nerve cord, though some authors refer to the combined structure as the nerve ring.) Nine neurons (RIR, RIV, RMDD, RMD and RMDV) are intrinsic to the nerve ring itself. An additional 26 neurons (AIA, AIB, AIM, AIN, AIY, AIZ, RIA, RIB, RIC, RIH, RIM, RIS, RMF and RMH) are intrinsic to the combined structure, for a total of 35 intrinsic neurons in the brain.

It should be understood that this estimate has ‘error bars’ because of definitional ambiguities. Ten motor neurons (RMH, RMF and RMD) could arguably be removed from the list, as it is unclear whether motor neurons qualify as intrinsic neurons. Or the brain could be enlarged by moving the posterior border further behind the excretory pore, which would add 10 neurons (RIF, RIG, RMG, ADE and ADA). To make these ambiguities explicit, we estimate 35 ± 10 intrinsic neurons. Of the 302 CNS neurons, 180 make synapses in the brain. Therefore, neurons intrinsic to the brain make up about 15 to 25% of brain neurons, and 8 to 15% of CNS neurons.

Skeletonization and path length calculation

We generated skeletons for all neurons marked as proofread using skeletor (https://github.com/navis-org/skeletor), which implements multiple skeletonization algorithms such as TEASAR. In brief, neuron meshes from the exported segmentation (LOD 1) were downloaded and skeletonized using the wavefront method in skeletor. These raw skeletons were then further processed (for example, to remove false twigs and heal breaks) and produce downsampled versions using navis (https://github.com/navis-org/navis). A modified version of this skeletonization pipeline is implemented in fafbseg (https://github.com/navis-org/fafbseg-py).

Quantifying cell volume and surface area

We calculated cell volumes and surface areas using CAVE’s L2Cache. Volumes were computed by counting all voxels within a cell segment and multiplying the count by the voxel resolution. Area calculations were more complicated and were performed by overlap through shifts in voxel space. We shifted the binarized segment in each dimension individually and extracted the overlap of false and true voxels. For each dimension, we counted the extracted voxels and multiplied the count by the voxel resolution of the given dimensions. Finally, we added up per dimension area estimates. This measurement will overestimate area slightly but smoothed measurements are ill-defined and were too compute intensive.