Claims 27
From the paper Measuring Stimulus-Evoked Neurophysiological Differentiation in Distinct Populations of Neurons in Mouse Visual Cortex · claims file · checked 2026-10-22: ready
- session-link
The dandiset's sessions can be matched to the 44 sessions the paper analysed (the results/main.parquet of the paper's code), and matched sessions agree on mouse, Cre line, area, depth and number of cells, and give the paper's per-trial ND.
How it will be judged
Added by the author of attempt 000036-mayner-2022-01. Sessions are matched on mouse id and start time, the NWB session_start_time minus 7 hours being equal to the start_time of main.parquet to the millisecond (the session identifiers differ; the offset was found while preparing the attempt). The link holds (reproduced) when all 44 paper sessions match one dandiset session each, every matched session agrees on Cre line, area, depth and the number of cells with a dF/F trace, and the dandiset's dF/F gives the paper's unnormalised per-trial ND to a relative 1e-6 for at least 90% of trials when the trial start may move by -8 to +3 frames from where the NWB stimulus times put it. Partially reproduced when all sessions match but an agreement holds for fewer sessions or fewer trials; not reproduced when fewer than 44 sessions match. The unmatched dandiset sessions are listed, and the one that fits the excluded L5 AL session is named.
- design-sessions
Data come from 45 two-photon sessions in nine mice (three per Cre line for L2/3, L4 and L5), 15 sessions per line and 9 per area over V1, LM, AL, PM and AM, with one L5 AL session excluded for recording problems.
Data were collected from L2/3, L4, and L5 in each of five areas (V1, LM, AL, PM, and AM; Fig. 1 E ) across 45 experimental sessions (15 sessions per transgenic line; 9 sessions per area; 5 ± 1 sessions per mouse ... One imaging session in L5 of AL was excluded from our analyses because of technical problems with the two-photon recording
Results, paragraph 1 and 2; Materials and Methods, paragraph 1How it will be judged
Counts of sessions, subjects, Cre lines and targeted areas in the dataset. Counts are exact (a tolerance of 10% is under one unit). Reproduced when the dataset holds 45 sessions from 9 mice, 3 mice and 15 sessions per line, 3 sessions per layer and area pair; a dataset that holds only the 44 analysed sessions, the one missing being in L5 AL, also counts as reproduced, since the paper excluded that session (Results, paragraph 1); partially reproduced when it holds a subset of them that matches these counts in part. The sex of each mouse (L2/3 two males, L4 three, L5 one) and sessions per mouse (5 +/- 1) are reported. Whether the excluded L5 AL session can be identified (by its absence from the code's results/main.parquet) is reported.
- ext1-2-cell-counts
4018 cells were recorded over the 44 analysed sessions, 1683 in L2/3, 1781 in L4 and 554 in L5, with per layer and area totals from 62 (L4 AM) to 649 (L4 V1).
How it will be judged
Per session, the number of cells (ROIs) whose dF/F traces the dataset holds, summed per layer, per layer and area pair, and overall, over the 44 analysed sessions. Reproduced when the totals are within tolerance; the 15 layer and area totals of table-s1.csv (L2/3 V1 451, LM 353, AL 336, PM 259, AM 284; L4 V1 649, LM 456, AL 479, PM 135, AM 62; L5 V1 126, LM 185, AL 86, PM 76, AM 81) are compared and the number within 10% reported. If the dataset holds more ROIs than the paper used (for example invalid ROIs), the count after any validity flag the dataset gives is shown too.
- ext1-3-kinetics-no-difference
Calcium indicator kinetics, measured as the half-life of an exponential fit to each cell's mean isolated-event-locked dF/F, do not differ across layers, areas or their interaction.
we found no layer × area interaction (likelihood ratio test; χ 2 (8) = 1.293, p = 0.996). We tested for main effects of layer and area in two further models and likewise found none
Extended Data Figure 1-3 legendHow it will be judged
A claim of absence, judged on the paper's tests. Per cell, isolated L0 events (no other event in the window around it), the mean dF/F trace locked to them, an exponential a0 + a1 exp(-lambda t) fitted by least squares from the event to the end of the window, half-life ln 2 / lambda (legend; code fig_1-3.ipynb). LME models of half-life with session as random intercept, fitted by maximum likelihood, layer * area against layer + area; then layer + area against area alone, and against layer alone (code). Reproduced when none of the three likelihood ratio tests has p < 0.05; not reproduced when any does. The window has readings (a) 50 ms before and 100 ms after (legend, main) and (b) 50 and 100 samples (code, variant). Events come from the dataset if it holds L0 events; otherwise from the linked l0_analysis.py run on the dataset's dF/F. The median R2 of the fits and the lowest R2 (0.61 in the paper) are reported.
- fig3a-layer-specific
Unscrambled stimuli evoke higher ND than their phase-scrambled versions in L2/3 but not in L4 or L5, with a significant layer by stimulus category interaction.
We fitted an LME model with stimulus category (unscrambled or scrambled), layer, and their interaction as fixed effects and found a significant interaction (likelihood ratio test, χ 2 (2) = 13.379, p = 0.00124). Post hoc tests showed that the unscrambled versus scrambled difference was specific to L2/3
Results, Increased differentiation for unscrambled stimuli is specific to excitatory cells in L2/3; Figure 3AHow it will be judged
ND as in the opening comment for the 5 stimuli of the contrast over the 44 sessions; the layer model and post hoc tests as in the opening comment (Methods, LME models and Post hoc tests; code stats.ipynb). Reproduced when the likelihood ratio test has p < 0.05, the L2/3 post hoc test has adjusted p < 0.05, and the L4 and L5 tests do not; partially reproduced when L2/3 is significant and the interaction or the absence in L4 or L5 fails. Cohen's d per layer over pooled trials (code fig_3 notebook, effect_sizes) is compared with the expected value; the post hoc z values and CI bounds are reported.
- fig3b-area-specific
The higher ND for unscrambled than scrambled stimuli is specific to areas AL and AM, with a significant area by stimulus category interaction.
We fitted an LME model with stimulus category, area, and their interaction as fixed effects and found a significant interaction (likelihood ratio test, χ 2 (4) = 15.203, p = 0.00430). Post hoc tests showed that the unscrambled versus scrambled difference was specific to AL and AM
Results, Increased differentiation for unscrambled stimuli is specific to areas AL and AM; Figure 3BHow it will be judged
As fig3a-layer-specific with area in place of layer. Reproduced when the likelihood ratio test has p < 0.05, the AL and AM post hoc tests have adjusted p < 0.05, and V1, LM and PM do not; partially reproduced when one of AL and AM is significant, or both are and another area is too. Cohen's d per area over pooled trials is compared with the expected values.
- ext3-1-natural-artificial
Over all 12 stimuli, naturalistic stimuli evoke higher ND than artificial ones in L2/3 and L5 (not L4) and in AL and AM (not V1, LM or PM), with significant interactions with layer and with area.
In this analysis, post hoc tests showed a significant effect also in L5; however, this contrast does not control for low-level stimulus characteristics and is thus harder to interpret
Extended Data Figure 3-1 legendHow it will be judged
As fig3a-layer-specific and fig3b-area-specific, on all 12 block stimuli (8 naturalistic, 4 artificial, as listed in the opening comment; 5280 trials; code stats.ipynb, Figure 3-1). Reproduced when both interactions have p < 0.05, L2/3 and L5 but not L4 are significant, and AL and AM but not V1, LM or PM are. The area chi2 is compared with the code's 22.213 and not with the legend's value, which repeats the layer test.
- ext3-2-sensitivity
The L2/3, AL and AM specificity of the unscrambled against scrambled ND difference holds for nearly all choices of distance metric, state length, frequency spacing and spectral window.
For nearly all other combinations of parameters, we likewise find that unscrambled stimuli elicit increased ND specifically in L2/3 of AL and AM
Extended Data Figures 3-2, 3-3 and 3-4 legends; Results, Increased differentiation for unscrambled stimuli is specific to areas AL and AM, paragraph 2How it will be judged
ND recomputed for the 108 parameter sets of the code (main.ipynb), with state length 0.2, 0.5, 1, 2, 5 and 10 s; Euclidean, city block and Chebyshev distance; linear or log-spaced frequency bins (code to_log_spacing); rectangular window without overlap (Fig. 3-2), Tukey (alpha 0.25) with 12.5% overlap (Fig. 3-3) and Kaiser (beta 14) with 50% overlap (Fig. 3-4), the windowed ones by scipy.signal.spectrogram rescaled to the FFT power as in spectral_differentiation.py. Then the layer and area models and post hoc tests of fig3a-layer-specific and fig3b-area-specific for each. A set agrees when both interactions have p < 0.05, L2/3 is the only significant layer and AL and AM are the only significant areas. The paper does not quantify "nearly all"; readings are at least 90% and at least 80% of the 108 sets. Reproduced when the share of agreeing sets meets the reading; the outcome is the one both readings give, and blocked as methods_underspecified when they differ. The share is also reported per window type, and for the looser rule that L2/3, AL and AM are each significant.
- ext3-5-arousal-covariates
With locomotion and pupil diameter as covariates in the LME models, L2/3 of AL and AM remain the populations where ND is greater for unscrambled than scrambled stimuli, for nearly all parameter combinations.
Consistent with the simpler models, for nearly all parameter combinations, L2/3 of AL and AM emerged as the cell populations in which ND is greater for unscrambled versus scrambled stimuli
Extended Data Figures 3-5, 3-6 and 3-7 legends; Results, Increased differentiation for unscrambled stimuli is specific to areas AL and AM, paragraph 3How it will be judged
As ext3-2-sensitivity with per-trial locomotion fraction and mean normalized pupil diameter added as fixed effects to the full and reduced models (code stats.ipynb, Figures 3-5 to 3-7). Same agreement rule and the same readings of "nearly all" (90% and 80% of 108 sets), crossed with the readings of artifact removal and pupil diameter named in the opening comment. The main parameter set (1 s, Euclidean, linear, rectangular) is reported on its own.
- ext3-8-events-spectral
Spectral ND of L0 calcium events gives results similar to dF/F, with greater ND for unscrambled than scrambled stimuli in L2/3 and in AL, while AM is at the threshold (p = 0.050).
Results are similar to the main analysis on ΔF/F 0 traces shown in Figure 3
Extended Data Figure 3-8 legendHow it will be judged
ND as in the opening comment applied to the per-frame L0 event magnitudes in place of dF/F (code main.ipynb, events). The paper's stats code asserts that all 2200 trials enter the models (stats.ipynb, EXPECTED_NUMBER_OF_OBSERVATIONS); only its plotting notebook drops trials whose log ND is minus infinity. If any trial has zero event ND, it is dropped (no log exists), the number dropped is reported, and this is recorded as a deviation. Events from the dataset if it holds L0 events; otherwise from the linked l0_analysis.py on the dataset's dF/F. Models and tests as in fig3a-layer-specific and fig3b-area-specific. Reproduced when both interactions have p < 0.05, L2/3 is significant and L4 and L5 are not, AL is significant and V1, LM and PM are not; the AM result (adjusted p near 0.05 in the paper) is reported and not judged.
- ext3-9-no-transients
With the 200 ms after each L0 event removed from dF/F, ND remains greater for unscrambled than scrambled stimuli in L2/3 and in AL, with significant interactions; AM is not significant (p = 0.0958).
Results are similar to the main analysis shown in Figure 3 , indicating that our ND results are not driven solely by initial transients in the calcium response
Extended Data Figure 3-9 legendHow it will be judged
For each cell, the 6 dF/F samples starting at each frame with an L0 event set to missing and filled by linear interpolation (legend; code main.ipynb, TAU = 0.2 s), then ND and the models of fig3a-layer-specific and fig3b-area-specific. Events as in ext3-8-events-spectral. Reproduced when both interactions have p < 0.05, L2/3 is significant and L4 and L5 are not, and AL is significant and V1, LM and PM are not; the AM result is reported and not judged.
- mediation-event-magnitude
Part of the effect of stimulus category on ND is mediated by mean calcium event magnitude and part is direct (mediated 0.1351, direct 0.0951, proportion mediated 0.59).
We also analyzed whether the mean magnitude of calcium events (a proxy for firing rate) mediated the effect of stimulus category and found evidence for both mediated and direct effects
Results, Increased differentiation for unscrambled stimuli is specific to areas AL and AM, paragraph 4How it will be judged
Tolerances are the distance from each value to the nearer bound of its 95% CI, so every value within tolerance also lies inside the paper's interval. Data are the 5 contrast stimuli in the L2/3 AL and AM sessions only (6 sessions, 300 trials; code stats.ipynb, the subset the Methods do not name). Mediator per trial is the mean over cells and frames of the L0 event magnitudes (code main.ipynb, mean_magnitude); covariates are per-trial mean normalized pupil diameter and locomotion fraction. Mediator model mean_magnitude ~ category + pupil + locomotion + (1 | session) and outcome model log10 ND ~ mean_magnitude + category + pupil + locomotion + (1 | session), both by maximum likelihood, and R mediation::mediate with treatment category (Methods, Mediation analyses), 1000 quasi-Bayesian simulations (the package default), robustSE = TRUE and seed 803317503 as in the code. Reproduced when all three values are within tolerance and both the mediated and the direct effects have p < 0.05; partially reproduced when both effects are significant and a value falls outside. Readings of artifact removal and pupil diameter as in the opening comment.
- table1-permutation
Per-session permutation tests find ND greater for unscrambled than scrambled stimuli in all 6 sessions of L2/3 AL and AM, in 8 of 15 L2/3 sessions, 1 of 15 L4 and 0 of 14 L5 sessions.
In all sessions recorded from L2/3 of AL and AM, responses to unscrambled stimuli were significantly more differentiated than to scrambled stimuli ( p < 0.05)
Table 1; Results, Permutation tests for individual experimental sessionsHow it will be judged
Counts are exact, since 10% of these counts is under one session. Per session, the difference in mean per-trial ND between the 20 unscrambled and 30 scrambled trials, with 20,000 random permutations of the category labels (Methods, Permutation tests). ND is log10 as in the code (table-1.ipynb); untransformed ND is a variant. p is the fraction of permuted differences greater than the observed (Methods) and, as a variant, greater than or equal with p <= 0.05 (code); significance at 0.05. Reproduced when all five counts match under the Methods rule; partially reproduced when the L2/3 AL and AM counts match and another differs by one session. The full 15-cell table is reported.
- fig4-locomotion-effect-size
In L2/3 of AL and AM, the per-session effect size of unscrambled against scrambled ND is positively correlated with the fraction of time spent running (r = 0.896).
We found that in L2/3 of AL and AM, effect sizes were positively correlated with locomotion activity (Pearson’s r = 0.896; two-sided t test; t (4) = 4.030, p = 0.0157, 95% CI [0.308, 1.00]
Results, Arousal is correlated with effect size; Figure 4, top leftHow it will be judged
Tolerance is the distance to the nearer CI bound (the interval is [0.308, 1.00]). Per session, Cohen's d of per-trial log10 ND, 20 unscrambled against 30 scrambled trials, pooled SD (code fig_4.ipynb). Locomotion fraction is the share of all the session's 2p frames with |filtered velocity| >= 2.5 cm/s (Methods, Locomotion; code behavior_by_session). Pearson correlation over the 6 L2/3 AL and AM sessions, two-sided. Reproduced when r is within tolerance and p < 0.05 under both readings of artifact removal; partially reproduced when r is positive and within tolerance but p >= 0.05. Blocked as methods_underspecified when the readings disagree. The per-session values and the correlation over all 44 sessions (Fig. 4, bottom) are reported.
- fig4-pupil-effect-size
In L2/3 of AL and AM, the per-session effect size is positively but not significantly correlated with mean normalized pupil diameter (r = 0.716, p = 0.109).
and pupil diameter ( r = 0.716; t (4) = 2.054, p = 0.109, 95% CI [–0.227, 1.00]
Results, Arousal is correlated with effect size; Figure 4, top rightHow it will be judged
Tolerance is the distance to the nearer CI bound. Effect size as in fig4-locomotion-effect-size; pupil diameter per frame, normalized by its maximum within the 10 blocks, averaged over all the session's 2p frames (code). Reproduced when r is positive and within tolerance under every reading of pupil diameter (Methods semi-major axis; code 2*sqrt(area/pi) as a variant) and artifact removal; blocked as methods_underspecified when they disagree. The p-value is reported.
- discussion-session-effect-sizes
Within single L2/3 AL and AM sessions effect sizes are medium to very large (Cohen's d 0.57 to 1.25), while the effect over these sessions pooled is 0.34.
while we observed medium to very large effect sizes within individual experimental sessions in L2/3 of AL and AM (Cohen’s d = 0.57–1.25), the overall effect was relatively subtle (Cohen’s d = 0.34)
Discussion, paragraph 6How it will be judged
Per-session Cohen's d as in fig4-locomotion-effect-size; the pooled d over all 300 trials of the 6 sessions with pooled SD (code fig_4.ipynb). Reproduced when the six per-session values lie in the range widened by 10% of each bound (0.513 to 1.375), the extremes are within tolerance, and the pooled d is within tolerance.
- fig5-multivariate-layer
Multivariate ND (mean distance to the centroid in an 8-dimensional UMAP embedding) is higher for unscrambled than scrambled stimuli specifically in L2/3.
The mean centroid distance was higher in response to unscrambled compared to scrambled stimuli ( Fig. 5 ), and this effect was specific to L2/3
Results, Multivariate analysis also shows increased differentiation for unscrambled stimuli; Figure 5AHow it will be judged
Per session, the dF/F of all cells over every 2p frame of the 50 trials of the 5 contrast stimuli, as a frames by cells matrix, reduced to 8 dimensions by UMAP with the parameters of umap-params.yml (n_neighbors 50, min_dist 0.5, Euclidean, spectral init, random_state 42; Methods, Multivariate differentiation). Per trial, the mean Euclidean distance of its frames to their centroid, log10 (code main.ipynb). Models and tests as in fig3a-layer-specific. Reproduced when the interaction has p < 0.05, L2/3 is significant and L4 and L5 are not. UMAP results depend on the library version; the version used is reported.
- fig5-multivariate-area
Multivariate ND is higher for unscrambled than scrambled stimuli specifically in AL and AM.
and areas AL and AM [area × stimulus category interaction ... likelihood ratio test, χ 2 (4) = 16.232, p = 0.00272
Results, Multivariate analysis also shows increased differentiation for unscrambled stimuli; Figure 5BHow it will be judged
Mean centroid distance as in fig5-multivariate-layer; the area model and tests of fig3b-area-specific. Reproduced when the interaction has p < 0.05, AL and AM are significant and V1, LM and PM are not.
- ext5-1-events-multivariate
Multivariate ND of L0 events is also higher for unscrambled than scrambled stimuli in L2/3 and in AL and AM, and in V1 as well.
The results of multivariate differentiation analysis of these data were consistent with our results using ΔF/F 0 traces (Extended Data Fig. 5-1 ), with the additional finding of significantly greater differentiation for unscrambled versus scrambled stimuli in L2/3 of V1 as well as AL and AM
Results, Multivariate analysis also shows increased differentiation for unscrambled stimuli, paragraph 3; Extended Data Figure 5-1How it will be judged
L0 event magnitudes summed in 1-s bins within the 5 contrast stimuli (Methods, Multivariate differentiation), then UMAP and mean centroid distance as in fig5-multivariate-layer, and the layer and area models of fig3a-layer-specific and fig3b-area-specific. Bins with no events are discarded before UMAP (Methods, main reading); the code keeps them (variant). Events as in ext3-8-events-spectral. The area model pools layers, so "L2/3 of V1" is tested as V1 in the area model and L2/3 in the layer model. Reproduced when both interactions have p < 0.05, L2/3 is the only significant layer, and V1, AL and AM are significant while LM and PM are not; partially reproduced when L2/3, AL and AM hold and V1 does not.
- fig6-category-decoding
Stimulus category (unscrambled or scrambled) can be decoded accurately from most layers and areas, well above chance, without the layer and area specificity of ND.
Decoding performance was high for most areas and layers ( Fig. 6 ), in contrast to the unscrambled-scrambled difference in ND
Results, Decoding analysis does not reveal layer or area specificity; Figure 6How it will be judged
Per session, the 50 trials of the 5 contrast stimuli, each trial the dF/F of all cells over its frames flattened (trials truncated to the shortest), PCA keeping 99% of variance and shrinkage LDA (solver lsqr, shrinkage auto) in one pipeline, 5-fold stratified cross-validation without shuffling, mean balanced accuracy over folds (Methods, Decoding analyses; code main.ipynb). Chance is 0.5. The paper does not quantify "high" or "most"; "most" is read as more than half of the 15 layer and area populations, and "high" as a mean accuracy over the population's sessions of at least 0.75 and, second reading, at least 0.9. Reproduced when, under both readings, most populations are high and every layer and every area has a mean accuracy above 0.5; blocked as methods_underspecified when the readings disagree. Per-session accuracies are reported. The paper reports no test of differences in decoding across layers or areas ("did not vary substantially", Abstract; "remarkably similar", Discussion), so the absence of specificity is shown, as accuracies by layer, area and pair, and is not judged.
- ext6-1-identity-decoding
Stimulus identity among all 12 stimuli can be decoded accurately from most layers and areas (chance 1/12).
Stimulus identity can be accurately decoded from most layers and areas using responses to all 12 stimuli
Extended Data Figure 6-1 legend; Results, Decoding analysis does not reveal layer or area specificityHow it will be judged
As fig6-category-decoding with all 120 trials of the 12 stimuli and stimulus identity as the class (code main.ipynb). Chance is 1/12. Same readings of "most" and "high" (mean balanced accuracy at least 0.75, and at least 0.9); reproduced when both readings agree that most populations are high and every layer and area is above 1/12.
- fig7-stimulus-effect
ND differs among the five continuous naturalistic clips, over all layers and areas.
We fitted an LME model with stimulus as a fixed effect and found it was significant (likelihood ratio test, χ 2 (4) = 32.115, p = 1.812e–6)
Results, Differences in ND among individual stimuliHow it will be judged
Per-trial log10 ND of the snake, crickets, man writing, mousecam and conspecifics clips over all 44 sessions; log10 ND ~ stimulus + (1 | session) against log10 ND ~ 1 + (1 | session), maximum likelihood, likelihood ratio test (Methods, LME models; code fig_7,7-1.ipynb). Reproduced when p < 0.05 and chi2 is within tolerance; partially reproduced when p < 0.05 only.
- fig7-snake-highest
The snake (predator) clip evokes higher ND than the conspecifics, crickets and man writing clips, and the mousecam clip higher ND than man writing.
revealed that the predator stimulus (a snake) evoked significantly higher differentiation than clips of conspecifics ... prey (crickets ... and a man writing ... The “mousecam” clip of movement through a wooded environment also evoked significantly higher differentiation than the clip of a man writing
Results, Differences in ND among individual stimuli; Figure 7How it will be judged
Pairwise differences of estimated marginal means from the model of fig7-stimulus-effect, emmeans(model, pairwise ~ stimulus) with Tukey adjustment, two-sided (Methods, Post hoc tests; code). The CIs are of the difference in log10 ND. Cohen's d of pooled per-trial log10 ND per pair (code). Reproduced when the four named pairs have adjusted p < 0.05 in the stated direction and the other six pairs do not (the Extended Data Fig. 7-1 legend speaks of "the four stimulus pairs identified as having significantly different levels of ND"); partially reproduced when three or more of the four hold. The estimated differences are checked against the CIs and reported.
- ext7-1-arousal-mediation
Arousal alone does not account for the differences among continuous clips, with a mixture of direct and arousal-mediated effects; for the largest contrast, snake against man writing, there is evidence only for a direct effect of stimulus on ND, with locomotion or pupil diameter as mediator.
Notably, for the largest contrast (predator vs man writing), for both locomotion and pupil diameter we found evidence only for a direct effect
Extended Data Figure 7-1 legendHow it will be judged
Per-trial data of fig7-stimulus-effect. For each arousal variable as mediator with the other as covariate, mediator model mediator ~ stimulus + covariate + (1 | session) and outcome model log10 ND ~ mediator + stimulus + covariate + (1 | session), maximum likelihood, then mediation::mediate with treat.value snake and control.value man writing (Methods, Mediation analyses; code fig_7,7-1.ipynb, seed 803317503, 1000 simulations; the code negates pupil diameter, which changes only signs). Reproduced when, for both mediators, the direct effect has p < 0.05 and the mediated effect does not; partially reproduced when this holds for one. The other three significant pairs are reported. Readings of artifact removal and pupil diameter as in the opening comment.
- fig8-nd-sd-correlation
Within L2/3 of AL and AM, mean ND per stimulus is positively correlated with stimulus differentiation across the 12 stimuli, but the noise stimulus is highly influential (Cook's D 2.318).
Within L2/3 of AL and AM, the mean ND elicited by each stimulus was positively correlated with SD (Pearson’s r = 0.746, one-sided t test; t (10) = 3.542, p = 0.00267, 95% CI [0.393, 1.00]
Results, SD does not explain ND; Figure 8How it will be judged
Tolerance of r is the distance to the nearer CI bound. SD per stimulus is ND as in the opening comment computed on the movie's pixel intensities (each pixel a cell, 30 frames/s, 1-s windows, Euclidean, median, no division by pixel count), after blurring each frame with a Gaussian whose half width at half maximum is the 8.92-degree RF radius (sigma per axis = radius in pixels / sqrt(2 ln 2), screen 120 x 95 degrees over 192 x 120 pixels, reflect mode; Methods, Spectral differentiation; code stimulus_properties.ipynb), then log10. Mean ND per stimulus is the mean of per-trial log10 ND over all trials of the 6 L2/3 AL and AM sessions (code fig_8.ipynb). Pearson correlation, one-sided (greater). Cook's D from the linear fit of mean ND on log SD. Reproduced when r is within tolerance with p < 0.05 and noise has the largest Cook's D, within tolerance. Movies from the dataset if it holds the stimulus frames; otherwise from the Zenodo deposit, a variant to record.
- fig8-no-sd-relation-controlled
Without the noise stimulus the ND and SD correlation is weak and not significant (r = 0.258), and among the scrambled stimuli and their unscrambled counterparts there is no evidence of a relation (r = -0.378); so ND is not inherited from SD.
If we exclude this stimulus, we find a weaker correlation ( r = 0.258; one-sided t test; t (9) = 0.801, p = 0.222, 95% CI [–0.307, 1.00] ii ). Furthermore, there was no evidence of a relationship with ND when considering only the scrambled stimuli and their unscrambled counterparts
Results, SD does not explain ND; Figure 8 legendHow it will be judged
A claim of absence, judged on the paper's tests with the values of fig8-nd-sd-correlation. Pearson correlation of mean log10 ND on log10 SD over the 11 stimuli other than noise, one-sided (greater), and over the 5 contrast stimuli, two-sided. Reproduced when neither has p < 0.05; not reproduced when either does. Both r values are reported against the paper's (0.258 and -0.378).
- ext8-1-no-low-level-relation
ND in L2/3 of AL and AM is not explained by stimulus luminance, contrast or spectral energy.
We also did not find a relationship with stimulus luminance, contrast, or spectral energy
Extended Data Figure 8-1 legend; Results, SD does not explain NDHow it will be judged
A claim of absence. Per stimulus, mean luminance is the mean pixel intensity and contrast the SD of pixel intensities of the unblurred movie; spectral energy is the sum of the energy spectral density of each pixel's intensity time series of the blurred movie after removing the DC component (legend; code stimulus_properties.ipynb). The paper names no test for this figure; readings are Pearson correlations of mean log10 ND (as in fig8-nd-sd-correlation) with each property over the 12 stimuli, (a) two-sided and (b) one-sided (greater), the tests of Fig. 8. DC removal has readings per-pixel mean (legend) and global mean (code). Reproduced when no correlation has p < 0.05 under any reading; not reproduced when one does under every reading; blocked as methods_underspecified when readings differ. All r values are reported.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none