Human-reviewed synapse samples used to validate the synapse classifier in Methods §"Synapse detection evaluation" and ED Fig. 1b–d.
| File | What it is |
|---|---|
2024-09-20_aelysia_synapse_sample_complete.csv |
The first round of Aelysia-team manual synapse review (Sep 2024) — completed sample. |
251013_evaluation_links.csv |
The 13 Oct 2025 evaluation batch — one row per to-review synapse, with the Neuroglancer link and the reviewer's eventual call. |
251013_synapse_evaluation.csv |
The 13 Oct 2025 per-synapse outcomes (the evaluator's verdict on each candidate). |
banc_888_v2_synapse_sample_small_2026-04-21.csv |
The v888 / synapses-v2 small-sample evaluation used in the paper. |
banc_888_v3_synapse_sample_2026-05-14.csv |
The v888 / synapses-v3 sample — used to compare v2 and v3 predictions on the same neurons. |
Per-synapse rosters were exported from CAVE; the verdict columns were
filled by hand by reviewers (Aelysia team for the 2024 round; the BANC
authors for the 2025/2026 rounds). The exact link between a candidate
synapse and its Neuroglancer view is the connector_id / coordinate
columns.
R/figures/panels_synapse_review.R reads these files to compute the
per-synapse precision / recall / sample-size tallies that go into ED
Fig. 1b–d, and writes data/synapse_capture/-style sidecars consumed by
R/text/numbers.R (e.g. synapse_v2_sample_fiveorless_* variables).