Dandiset 001453 · Reproduction registry

Claims

From the paper Neural circuits underlying divergent visuomotor strategies of zebrafish and Danionella cerebrum

Reproducing Fouke et al. (2025) from dandiset 001453 10

Notebook reproduce_fouke_2025.ipynb

Not reproducedResults, first section; Figure 1D legendnames its codes from outside the dandiset
DC lack the direction-dependent velocity modulation seen in ZF.
By the paper's test (Tukey's HSD over the 18 species and stimulus groups) 3 of 36 DC pairs differ, all static against a motion direction (backward left, backward right, forward left; adjusted p 0.030 to 0.049). The DC static values come from 9 fish and include one at 106.5 mm/s and two near 0. Among the eight directions no DC pair differs (smallest adjusted p 0.999), DC per-stimulus medians span 8.81 to 10.52 mm/s (ratio 1.19, ZF 2.03), and a one-way ANOVA across the nine DC stimuli gives p = 0.479 (ZF p = 0.00082). With values above 100 mm/s left out, no DC pair differs. The same Tukey test finds no ZF pair over the 18 groups (smallest p 0.835). The printed output of Fig1-S1.ipynb finds the same three DC pairs significant (adjusted p 0.0236, 0.0373 and 0.0397), so the paper's own test disagrees with the claim and this result matches the paper's computation. The DC-sub-2 static value of 106.5 mm/s is one trial measured over a 14.6 s motion phase in a 100 mm dish, likely a tracking artefact. It passes the code's speed check because the check covers 0 to 35 s (21.9 mm/s from the stored `dist`, 92.5 from positions); computing the check from positions, or leaving out zero medians as the code does, gives the same 3 pairs. The outcome is a property of the paper's test on this data, not evidence that DC modulate their speed with stimulus direction.
How the test differed from the paper’s method
- The DC static block is the code's window, but the code's filter on DC `vel` cannot be applied (DC static 10.52 against the code's printed 11.9 mm/s).
Not reproducedResults, second section; Figure 1H (Kruskal-Wallis test with Bonferroni correction)names its codes from outside the dandiset
Across directional stimuli ZF have higher angular acceleration than DC, and the distributions of the two species differ for every stimulus.
With angular acceleration as the Methods define it, DC are higher than ZF for all nine stimuli (medians 143 to 191 against 55.4 to 66.4 deg/s2, Kruskal-Wallis p below 0.05/9 for each), the reverse of the claim. Per-fish medians, and the files at 128 to 171 Hz alone (ZF 84.9, DC 178.6 deg/s2, p = 6.3e-05), give the same. The quantity Fig1-S1.ipynb computes gives ZF higher with p below 0.05/9 for 7 of 9 stimuli; it divides the heading change per frame by the file's mean frame interval, which in DC files is a median 8.73 times the frame interval (ZF 1.37), as DC files hold only trial frames. Part of the DC values may be tracking noise. In the 3 s stationary phase the DC median is 205 deg/s2, higher than the 171 during motion (per fish median ratio 1.24, p = 0.016), but ZF are also higher at rest (66.9 and 60.8, ratio 1.07, p = 0.00012), so this does not separate the species. In frames where the fish is still (below 1 mm/s) DC are higher than ZF (per-fish medians 72.4 and 34.9 deg/s2, Mann-Whitney p = 0.00014), and DC are higher in every speed band (at 5 mm/s or more, 207 and 85.8). The outputs cannot say whether the higher values in moving DC are noise or turning during continuous swimming. With the heading from the unwrapped `f0_theta` DC are higher still (218 to 278 against ZF 61.8 to 77.3 deg/s2, 0 of 9). The reversal should not be read as evidence against the paper.
How the test differed from the paper’s method
- Follows the Methods' definition (degrees per second squared) on heading resampled to 30 Hz; the code's quantity, which disagrees with the Methods, is shown as a variant.
- Heading is taken from the cumulative sum of `f0_vtheta`, as the code does; in DC files it does not match `f0_theta`.
Partially reproducedFigure 1G; Results, second sectionnames its codes from outside the dandiset
Both species follow the direction of optic flow, turning right (positive angular velocity) to rightward motion and left to leftward motion.
ZF turn with the motion; the rightward minus leftward difference is positive in 23 of 25 fish (median 29.51 deg/s, one-sided Wilcoxon p = 9.8e-07). DC do not by this test; 7 of 12 positive, median 0.88 deg/s, p = 0.151, and the DC median for left motion is +0.25 deg/s, the same sign as the paper's code printed (+0.3). DC turn right to right motion (8.67 deg/s).
How the test differed from the paper’s method
- The paper gives no test for following the direction; the per-fish rightward minus leftward difference and its test were chosen here before the analysis.
ReproducedData and Code Availability; the paper's Fig1-S1.ipynb at GitLab commit 33a5c29names its codes from outside the dandiset
The dandiset's stimulus codes can be linked to the stimuli the paper names, using the paper's analysis code and the protocol file of its stimulus software, and the values computed through that link agree with those the paper's code printed.
Codes are matched by their integer value. All 21 codes are in the dandiset, including every code the analysis uses. Through the link, 17 of 18 species and stimulus speed medians are within 10% of the values printed by Fig1-S1.ipynb (DC static is 10.52 against 11.9 mm/s), and the angular velocities differ from the printed ones by a median of 0.02 deg/s. The fish counts (25 ZF, 12 DC) match the code's lists; fish cannot be matched one by one, as no identifier links the dandiset's subjects to the code's folders. The static window is the code's (first 35 s of the code-20 block, measured from 3 s), and no fish median is exactly 0, so the code's exclusion of zero medians changes nothing. On codes 9 and 11 the ZF medians are 0.02 and 1.59 deg/s. Over 3 to 35 s DC turn more to the right for code 11 than for code 9 (median difference 5.82 deg/s, 10 of 12 fish, Wilcoxon p = 0.0093), and ZF go the same way without significance (1.66 deg/s, 16 of 25 fish, p = 0.141). If fish turn with lateral motion as they do with medial motion, code 11 is lateral right, as the protocol file has it; the data do not show that they do, so the assignment is not settled.
How the test differed from the paper’s method
- The code's filter on a `vel` column is not applied to DC trials, as the DC files have no such series; this is the remaining difference that may explain the one DC static value outside 10%.
- The timing of the DC stimulus labels could not be confirmed from the turning response (the contrast is 4.6 deg/s with true labels and 0.1 with shifted ones); the DC labels are taken as they are.
ReproducedResults, first section; Figure 1D
Danionella cerebrum (DC) swim at a higher median velocity than zebrafish (ZF) across the visual stimuli.
Over all 21 codes, the median over fish and code values is 94.0 pixels/s for DC and 55.2 for ZF; species effect F = 78.4, p = 6.3e-18. The ANOVA treats each fish and code value as independent; with one value per fish, DC 87.4 against ZF 52.9 pixels/s, Mann-Whitney p = 0.02. With positions resampled to 30 Hz, 95.2 against 53.8 (p = 1.3e-08). DC are faster within each frame-rate group.
How the test differed from the paper’s method
- Tested over all stimulus codes without their names; the paper's figure uses nine whole-field stimuli.
- The comparison assumes one spatial scale for all rigs, which the files do not state.
ReproducedResults, first section; Figure 1Dnames its codes from outside the dandiset
Across the whole-field stimuli DC have a higher median velocity than ZF (9.8 against 6.4 mm/s), with a significant main effect of species in a two-way ANOVA.
All four readings put both values within tolerance; (a) median of stimulus medians DC 10.02 and ZF 6.39 mm/s, (b) median of fish and stimulus values 9.87 and 6.41 (given as measured), (c) median of fish medians 9.36 and 6.33, (d) median of trials 9.88 and 6.28. Species effect F = 20.8, p = 7.5e-06. With 30 Hz positions DC 9.43 to 10.15 and ZF 6.07 to 6.30 (p = 0.0011). Without the trial rejection (4487 trials in place of 4458) readings (a) to (c) are unchanged and (d) gives DC 9.85 and ZF 6.30 (p = 6.7e-06).
How the test differed from the paper’s method
- The paper does not say how the overall median is taken; four readings are shown, and all match.
- 25 ZF, where the paper states 26.
- Speeds rely on the code's scale of 10 pixels per mm, which the dandiset does not state.
ReproducedResults, first section; Figure 1Dnames its codes from outside the dandiset
ZF modulate their velocity with the direction of motion; forward motion increases their median velocity (static 5.0 mm/s, forward 8.8 mm/s).
ZF static 4.84 mm/s over 18 fish and forward 8.78 mm/s over 25 fish. Of 18 fish with both, 14 are faster to forward motion (median difference 3.67 mm/s, Wilcoxon p = 0.0077). The static value is 8.31 mm/s in the 5 fish from 167 to 171 Hz files and 4.50 in the 13 from 32 to 35 Hz files. Static is the start of the blank block after all motion trials (4 hours in the protocol, 0.2 to 787.3 min in the files), one period per fish, and 7 ZF have none.
How the test differed from the paper’s method
- 25 ZF, where the paper states 26.
ReproducedFigure 1D legendnames its codes from outside the dandiset
DC swim at a significantly higher velocity than ZF for oblique stimuli.
Under reading (b), motion at 90 degrees or more, DC are significantly faster than ZF for right (p = 0.0002) and backward right (p = 0.0047), 2 of 4; under reading (a), the intercardinal directions, for 1 of 4 (backward right). Backward motion also differs (p = 0.0002).
How the test differed from the paper’s method
- The legend's "oblique" can be read two ways; the claim is reproduced under one of them (b) and not under the other (a).
ReproducedResults, second section; Figure S1Enames its codes from outside the dandiset
DC generally respond less than ZF to monocular stimuli, suggesting that they need more visual motion input to drive the optomotor response.
Species effect F = 23.3, p = 2.5e-06. DC turn less toward the stimulus than ZF for medial stimuli (3.26 against 13.48 deg/s, p = 0.0003) and binocular ones (2.87 against 21.87, p = 0.0002). For lateral stimuli DC are below ZF only nominally (-1.07 against -0.13 deg/s, p = 0.96); neither species turns toward lateral motion. With the protocol file's assignment of codes 9 and 11, the lateral medians change sign and DC are no longer below ZF there, which would make the claim partially reproduced. The DC turning to codes 9 and 11 leans toward the protocol file's assignment (see stimulus-code-link), so the lateral part is fragile.
How the test differed from the paper’s method
- Codes 9 and 11 are assigned as the figure code has them; the stimulus software's conversion has them the other way, and the data do not settle it.
ReproducedResults, first section; Figure 1C and Figure S1D
ZF swim in discrete bouts separated by glides, while DC swim continuously at a relatively constant velocity.
The median-to-mean ratio of instantaneous speed is 0.306 in ZF (25 fish, 0.244 to 0.580) and 0.908 in DC (12 fish, 11 above 0.74, one at 0.140), Mann-Whitney p = 5.4e-05. It holds in files at 128 to 171 Hz alone (0.298 against 0.898, p = 0.0028) and in forward trials alone.
How the test differed from the paper’s method
- The paper shows this with representative traces and a bout analysis whose bout files are not in the dandiset; the measure was chosen here before the analysis.

Not tested in a notebook 10

1 metadata insufficient, 9 data missing from the dandiset

BlockedResults, first and second sections; Figure 1C, 1E and 1F· Metadata insufficient
In representative trials ZF bouts reach a peak velocity of 44.7 mm/s with a median of 1.8 mm/s, DC swim at a median of 13.7 mm/s to forward motion; to oblique motion ZF show a median angular velocity of 12.0 degrees/s and median peak angular acceleration of 174 degrees/s2, and DC an average peak of 103 degrees/s2 to rightward motion.
The numbers come from single representative fish and trials. Fig1-S1.ipynb selects them by the index of a data folder and a trial number (for example DC folder 29, trial 79), and the dandiset gives no mapping from those folders to its subjects, nor the original file names.
BlockedFigure 2B to 2F· Data missing from the dandiset
Both species have spatially intermixed, direction-selective neurons enriched in the pretectum and anterior hindbrain, and share the major functional response types.
The two-photon imaging data are not in the dandiset, which holds only freely swimming tracking. The GitLab repository the paper names for "all behavioral and imaging data" holds notebooks only, and no other dandiset of this lab or paper was found on the archive.
BlockedFigure 2G· Data missing from the dandiset
The average direction selectivity index is similar in ZF and DC (0.62 against 0.66, p > 0.05, Student's t-test, 5 fish each).
Imaging data not in the dandiset, nor found elsewhere (see fig2d-shared-architecture).
BlockedFigure 2H· Data missing from the dandiset
ZF tend to have a higher proportion of motion-responsive neurons than DC (0.13 against 0.06, p = 0.059).
Imaging data not in the dandiset, nor found elsewhere (see fig2d-shared-architecture).
BlockedFigure 2K· Data missing from the dandiset
DC have a higher proportion of purely monocular neurons than ZF (0.37 against 0.16, p < 0.05).
Imaging data not in the dandiset, nor found elsewhere (see fig2d-shared-architecture).
BlockedFigure 2L· Data missing from the dandiset
The binocularity index ratio is higher in ZF than in DC (0.72 against 0.24, p = 0.010).
Imaging data not in the dandiset, nor found elsewhere (see fig2d-shared-architecture).
BlockedFigure 3E and 3F· Data missing from the dandiset
Head-fixed DC perform longer swims than ZF (median 0.75 s against 0.29 s, p < 0.0001).
The head-fixed tail tracking recorded during imaging is not in the dandiset, nor in the paper's GitLab repository.
BlockedFigure 3C and 3D· Data missing from the dandiset
Head-fixed ZF show larger directional tail angles than DC to turn-inducing stimuli, and shorten their swims for oblique stimuli while DC do not.
The head-fixed tail tracking is not in the dandiset, nor in the paper's GitLab repository.
BlockedFigure 4E and 4F· Data missing from the dandiset
Swim-onset-associated neurons in DC have a longer active (ON) phase than in ZF, while non-motor-associated neurons have similar ON durations.
Needs the imaging and the head-fixed tail tracking, neither of which is in the dandiset.
BlockedFigure 4H and 4I· Data missing from the dandiset
Motor-associated neurons in the putative ventromedial spinal projection neurons of both species encode turn direction in a lateralized way, correlating more with ipsilateral turns.
Needs the imaging, the tail tracking and the registration to the mapzebrain atlas, none of which is in the dandiset.

Findings about the data

What attempts found about the dandiset itself, for its owner and the archive.

Data missingThe dandiset holds only the freely swimming tracking of Figure 1 and Figure S1. The two-photon imaging and the head-fixed tail tracking behind Figures 2 to 4 are not in it, and the GitLab repository the paper names for all behavioural and imaging data holds analysis notebooks only. Nine of the paper's main findings cannot be tested.
Description missingThe stimulus codes 0 to 20 have no names or description in the files or the dandiset metadata. They can be named only from the paper's analysis notebook and the protocol file of its stimulus software.
Published code disagreesThe paper's analysis notebook labels code 9 as lateral right and 11 as lateral left, while its conversion of the stimulus software's names gives the reverse (it carries a remap that swaps them). The data do not settle which is right; zebrafish barely turn to either, and Danionella turn more to the right for code 11, which fits the stimulus software's assignment only if the fish turn with lateral motion.
Metadata errorSubject ids 0 to 11 are each used for one zebrafish and one Danionella cerebrum, so subject_id does not identify an animal. The archive's summary counts 25 subjects and lists only Danio rerio; the files hold 37 fish of two species.
Metadata errorf0_vtheta is stated in radians/s but holds the heading change per frame in radians, and f0_vx and f0_vy are stated in pixels/s but are per frame. In most Danionella files the sum of f0_vtheta over a trial does not match the change of f0_theta (median ratios 0.09 to 0.89).
Other18 of 25 zebrafish files and 1 of 12 Danionella files are sampled at 32 to 35 Hz, where the Methods state tracking at more than 150 Hz. Their timestamp intervals are regular (28.8 to 30.8 ms) and their trials last the protocol's 35 s, so these are the acquisition rates. Some measures differ with frame rate, such as the zebrafish static speed (8.31 against 4.50 mm/s).
Description missingNo spatial scale is given; positions are in pixels. They are consistent with the 10 pixels per mm in the paper's code, except one Danionella file (sub-5) whose positions reach 1490 pixels against about 1120 in the others.
OtherZebrafish and Danionella files were exported differently. Zebrafish files keep untracked frames and frames between trials; Danionella files hold only tracked trial frames and add a `dist` series. The paper's code filters Danionella frames on a `vel` column that is not in the files.
Published code disagreesThe paper's code computes the angular acceleration of Figure 1H as the heading change per frame divided by the file's mean frame interval, an angular velocity. In Danionella files the mean interval includes the gaps between trials (a median 8.73 times the frame interval), which scales their values down. With the Methods' definition Danionella have the higher angular acceleration.
Published code disagreesFigure 1H and its legend state a Kruskal-Wallis test with Bonferroni correction; the code runs two-sample Kolmogorov-Smirnov tests, and its loop selects the stimulus with a variable left from an earlier loop, so the nine tests compare the same stimulus.
Published code disagreesThe Results state that Danionella lack the direction-dependent speed modulation of zebrafish, but the Tukey test printed by Fig1-S1.ipynb for Figure 1D finds three Danionella pairs significant (static against backward left, backward right and forward left, adjusted p 0.024 to 0.040), as this attempt does.
OtherThe `dist` series of the Danionella files does not agree with the positions; over the first 35 s of a trial its sum is a median 1.11 times the path length of the smoothed positions, with a 5th to 95th percentile of 0.31 to 3.79. The paper's speed check uses it, though computing the check from positions passes the same trials.
Published code disagreesThe Figure 1D legend gives 26 zebrafish; the paper's code lists 25, its ANOVA degrees of freedom fit 25, and the dandiset holds 25.
Selection not documentedThe dandiset holds the 25 zebrafish and 12 Danionella the code keeps after exclusion (its comments give 28 and 40 recorded). The dandiset does not say that the excluded fish were left out or why.

Notebooks

Every registered entry that uses this dandiset. An entry can hold several notebooks, and can use more than one dandiset.

Reproduction Optomotor swimming of zebrafish and Danionella cerebrum (Fouke et al. 2025)

Tests the freely swimming findings of Fouke, He, Loring and Naumann (2025) on dandiset 001453, which holds the tracking data of 25 zebrafish and 12 Danionella cerebrum larvae during the optomotor protocol. It compares swimming speed, turning and angular acceleration between species and stimuli, the response to monocular stimuli, and how steady the swimming is. The imaging and head-fixed tail tracking of the paper's Figures 2 to 4 are not in the dandiset. All files

Reproducing Fouke et al. (2025) from dandiset 001453
reproduce_fouke_2025.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-14Agreed with 20 of 20 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The first executed notebook with the conclusion cells and the summary removed, the helper module and the claims file, with the paper's accepted manuscript (PMC12218792). The verifier listed the version's assets through the DANDI API and ran synthetic checks of angular_acceleration, trial_measures and the steadiness measure.
Round 2
2026-10-14Agreed with 20 of 20 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook). Reviewed: The second executed notebook in full, with the conclusion cells and the summary, the helper module, the claims file and the attempt record. The verifier did not download the paper again.

Method differences

How the tests differed from the paper’s method or data, and the claims each applies to.

25 ZF, where the paper states 26.fig1d-median-velocity fig1d-zf-forward-faster-than-static
The code's filter on a `vel` column is not applied to DC trials, as the DC files have no such series; this is the remaining difference that may explain the one DC static value outside 10%.stimulus-code-link
The timing of the DC stimulus labels could not be confirmed from the turning response (the contrast is 4.6 deg/s with true labels and 0.1 with shifted ones); the DC labels are taken as they are.stimulus-code-link
Tested over all stimulus codes without their names; the paper's figure uses nine whole-field stimuli.fig1d-dc-faster-all-codes
The comparison assumes one spatial scale for all rigs, which the files do not state.fig1d-dc-faster-all-codes
The paper does not say how the overall median is taken; four readings are shown, and all match.fig1d-median-velocity
Speeds rely on the code's scale of 10 pixels per mm, which the dandiset does not state.fig1d-median-velocity
The DC static block is the code's window, but the code's filter on DC `vel` cannot be applied (DC static 10.52 against the code's printed 11.9 mm/s).fig1d-dc-no-modulation
The legend's "oblique" can be read two ways; the claim is reproduced under one of them (b) and not under the other (a).fig1d-dc-faster-oblique
The paper gives no test for following the direction; the per-fish rightward minus leftward difference and its test were chosen here before the analysis.fig1g-follow-direction
Follows the Methods' definition (degrees per second squared) on heading resampled to 30 Hz; the code's quantity, which disagrees with the Methods, is shown as a variant.fig1h-zf-higher-angular-acceleration
Heading is taken from the cumulative sum of `f0_vtheta`, as the code does; in DC files it does not match `f0_theta`.fig1h-zf-higher-angular-acceleration
Codes 9 and 11 are assigned as the figure code has them; the stimulus software's conversion has them the other way, and the data do not settle it.figs1e-dc-respond-less-monocular
The paper shows this with representative traces and a bout analysis whose bout files are not in the dandiset; the measure was chosen here before the analysis.fig1c-continuous-vs-bouts

Provenance

Attempt record: 001453-fouke-2025-01
Paper: Neural circuits underlying divergent visuomotor strategies of zebrafish and Danionella cerebrum
Date: 2026-10-14
Made by: An agent, run by CatalystNeuro
Dandiset version: 0.250518.1950
Harness: Claude Code (GitHub Actions)
Model: claude-opus-5-5
Protocol: pilot-5

Transcripts of the agent runs

Each run’s prompt, messages and tool calls, with long output cut and the text of the paper left out, and its final report.

← Dandiset 001172 Dandiset 001528 →