Claims 34
From the paper Multi-timescale Rhythmic Dynamics in Rostral Ventromedial Medulla Neurons · claims file · checked 2026-10-21: ready
- dataset-composition
The evoked protocol recorded 104 cells in 53 animals (35 OFF, 45 ON, 24 NEUTRAL) and the ongoing protocol 62 cells in 30 animals (25 OFF, 25 ON, 12 NEUTRAL).
total of 104 cells were recorded in 53 animals: 35 OFF-cells, ... 45 ON-cells, and 24 NEUTRAL-cells ... we recorded 62 cells in 30 ... animals: 25 OFF-cells, 25 ON-cells, and 12 NEUTRAL-cells
Methods, Experiments, paragraph 4; Supplementary Table S11How it will be judged
Cells are units labelled ON, OFF or NEUTRAL; animals are subjects. The paper gives two groupings. Reading (a), the Methods: evoked is the "evoked" and "evoked/ongoing" files together (45 ON, 35 OFF, 4 + 20 NEUTRAL), ongoing is the "ongoing" files. Reading (b), Table S11: three sets, Evoked (45, 35, 4 in 35 animals), Evoked/Ongoing (0, 0, 20 in 18 animals), Ongoing (25, 25, 12 in 50 animals). Reproduced when, under one reading, every count per class and protocol is within tolerance (10% of the reported value). The exact counts are reported either way, with the number of subjects and sessions, since the two readings of the paper disagree on the animals.
- fig1c-psth-counts
The population PSTHs of Figure 1c pool 70 ON and 60 OFF cells, with 185 ON and 155 OFF trials aligned to withdrawal and 182 ON and 151 OFF trials aligned to heat onset, for windows that run to 150 s.
shows peri-stimulus time histograms (PSTHs) for 70 ON ... cells and 60 OFF-cells, aligned to the time of paw withdrawal ... In total, 185 ON-cell and 155 OFF-cell trials ... were aligned to withdrawal, while 182 ON-cell and 151 OFF
Results, Canonical ON/OFF responses to noxious stimulation, paragraph 3; Figure 1cHow it will be judged
All ON and OFF units of all files. The window ends 150 s after the event (Results, this paragraph); its start is open: (a) -30 s, as in Panel_1.ipynb (plot_spikes_around_event_fancy_figure, 0.1 s PSTH bins), (b) -10 s, the Methods' pre-event window (Data processing). Reproduced when the counts match under one reading. Trials kept by the rule of Spike2Tools.get_all_event_spike_trains, which applies the Methods' exclusion (Data processing, "if the recording ended within the window or if the next event ... truncated the usable interval"): for withdrawal alignment, the trial is dropped when the next heat onset, pinch or recording end is less than 150 s after the withdrawal, or the previous withdrawal or pinch (or the recording start) is less than 30 s before it; for heat alignment, the same with the next heat, pinch or end and any previous event (with 30 s or 10 s before the event, by reading). Reproduced when all six counts are within tolerance under one reading. The exact counts are reported for both.
- fig1c-withdrawal-locked
Population ON and OFF activity is sharply time-locked to paw withdrawal; aligned to heat onset the PSTHs show a clear lag and shallower slopes.
PSTHs showed a clear lag and shallower slopes when ... aligned to heat onset compared to withdrawal
Results, Canonical ON/OFF responses to noxious stimulation, paragraph 3; Figure 1c legendHow it will be judged
The paper quantifies the slopes with the piecewise models (Results, Evoked ON/OFF activity, paragraph 9; Tables S3, S5, S6, S7). It does not quantify the lag; the reading used is the switch times of the same fits, which the tables give: ON t_switch -0.184 s withdrawal-aligned (Table S3) against 8.694 s heat-aligned (Table S5); OFF 1.346 s (Table S7) against 11.718 s (Table S6). Using the four fits of fig2-trial-counts and tables3-s7-withdrawal-steeper: reproduced when, for both cell types, the heat-aligned t_switch 97% HDI lies wholly after the withdrawal-aligned one (the paper judges differences of these fits by their HDIs, as in tables3-s7-withdrawal-steeper), and tables3-s7-withdrawal-steeper is reproduced. Partially reproduced when only one of the two parts holds.
- fig2-trial-counts
For the models, windows from 10 s before to 100 s after the event gave 195 ON and 170 OFF trials aligned to withdrawal and 197 ON and 172 OFF trials aligned to heat onset; cells were not selected by pre-trial firing and every trial with a withdrawal was included.
This resulted in 195 ON-cell and ... 170 OFF-cell trials aligned to withdrawal, and 197 ON-cell ... and 172 OFF-cell trials aligned to heat onset
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 1How it will be judged
As fig1c-psth-counts, with the window -10 s to +100 s (Methods, Data processing; param_dict.json of each model notebook, PRE_EVENT_PERIOD -10, POST_EVENT_PERIOD 100). All ON and OFF units of the evoked, ongoing and evoked/ongoing files, as the notebooks load them (read_dataset_by_protocol with all three protocols). This trial set, binned at 50 ms (Methods, Piecewise Bayesian modeling; BIN_SIZE 0.05) and summed over all trials of a cell type into 2200 counts, is the input of every evoked model claim. Reproduced when all four counts are within tolerance; the exact counts and the number of cells contributing trials are reported.
- figs8-null-model-worse
A null model with a constant mean spike count fits significantly worse than every piecewise model, for ON and for OFF cells, so both an evoked departure and a recovery are needed.
Both null models clearly performed worse than either of the parametric fits.
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 2; Supplementary Figure S8How it will be judged
The null model is build_mean_model of bayesian_cell_models.py (notebooks 20 and 27), fitted to the same summed counts. Each model is sampled as the Methods give it (NUTS in PyMC, 4 chains, 1000 tuning steps, 1000 draws; target_accept 0.8 from sample_model) and compared by the expected log predictive density. The paper names ELPD without fixing the estimator: the model comparison plots use PSIS-LOO (plot_loo_comparison; notebook 23 prints LOO and WAIC, notebook 17 WAIC), so both are readings. Its word "significantly" is not defined; the readings are that the null model's ELPD is below each piecewise model's by more than (a) one or (b) two standard errors of the difference (dse of arviz.compare). Reproduced when, for each cell type, the null model is below every piecewise model of that type under one reading of each. The paper's notebooks print ELPD-LOO -11683 for the ON null against -7215 and -7454, and -9237 for the OFF null against about -6265 (difference 2972, dse 124).
- fig2g-switch-times
The switch from response to recovery comes 0.18 s before withdrawal for ON cells (97% HDI 0.27 to 0.095 s before) and 1.35 s after it for OFF cells (97% HDI 0.85 to 1.9 s after), so ON cells peak slightly before withdrawal while OFF cells keep falling after it.
differed for ON- and OFF-cells (Figure 2g) ... 0.18 s before withdrawal (97% HDI: 0.27–0.095 s before), and ... switch = 1.35 s after withdrawal (97% HDI: 0.85–1.9 s af ... Thus ON-cells peak slightly before withdrawal, while ... OFF-cells enter a brief post-withdrawal depression
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 3; Figure 2g and 2iHow it will be judged
The ON double-exponential model (build_double_exponential_on_model) and the OFF linear plus exponential model (build_linear_slope_recovery_off_model) fitted to the withdrawal-aligned counts of fig2-trial-counts, with the Methods' priors (Piecewise Bayesian modeling; the notebooks set the same values: ON log r ~ N(3, 0.5), log k_pre ~ N(4, 0.1), log k_post ~ N(2.5, 1), rates LogNormal(log 0.1, 0.5) and LogNormal(log 1/50, 0.5), t_s ~ U[-4, 4]; OFF b ~ HalfNormal(4), a and c = b + HalfNormal(4), k3 ~ HalfNormal(1), u0 ~ U[b, a], t_s ~ U[-4, 4]). The Methods give k, k2 ~ N(0, 1); the code places these priors on log k and log k2, which is run as a variant. Sampling as in figs8-null-model-worse. The tolerance is the larger distance from the reported value to its HDI bound; reproduced when both posterior means lie inside the paper's 97% HDIs and the ON mean is below 0 and the OFF mean above 0. The text's ON HDI (-0.27 to -0.095) differs from Table S3 (-0.257 to -0.108) and the OFF one (0.85 to 1.9) from Table S7 (0.904 to 1.832); the wider interval of each pair is used, and the disagreement is recorded. The Discussion (paragraph 3) puts the OFF nadir more than 2 s after withdrawal, which disagrees with the Results' 1.35 s; the Results' value is tested.
- fig2g-on-double-exponential
ON-cell recovery is better described by a double exponential, two decay timescales, than by a single exponential, which misses the sharp peak and the prolonged decay.
As shown in Figure 2c,g, ON-cell recovery was best captured ... by a double exponential model
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 4; Figure 2c, 2e and 2gHow it will be judged
The double-exponential and single-exponential ON models (build_double_exponential_on_model, build_single_exponential_on_model, priors as in their notebooks 18 and 19) fitted to the withdrawal-aligned ON counts; ELPD as in figs8-null-model-worse, readings PSIS-LOO and WAIC. The paper says "best captured" and states no margin, so, as in fig2h-off-linear-exponential, reproduced when the double exponential has the higher ELPD under either estimator. The difference and its standard error are reported. The paper's notebook 17 prints ELPD-WAIC -7215.5 against -7453.6, difference 238, dse 23.
- fig2c-on-time-constants
ON-cell firing rises before withdrawal with time constant 0.653 s and recovers with a fast component of 4.157 s and a slow component of 164.223 s.
constant τpre = 0.653 s (97% HDI: 0.570–0.739 s). After paw ... τfast = 4.157 s (97% HDI: 3.582–4.722 s) and a slow compo ... nent τslow = 164.223 s (97% HDI: 129.571–199.809 s)
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 4; Supplementary Table S3How it will be judged
The ON double-exponential fit of fig2g-switch-times; the time constants are the inverses of the rate parameters (tau = 1/beta), as in the Methods and notebook 18. Reproduced when the three posterior means lie inside the paper's 97% HDIs (the tolerance is the larger distance to an HDI bound); partially reproduced when two do.
- fig2h-off-linear-exponential
OFF-cell recovery is best described by the combined linear and exponential model; the double-exponential alternative fits the early recovery but fails to reach baseline, and its ELPD is only slightly lower.
For OFF-cells, recovery was best described by the combined ... linear and exponential model (Figure 2d,h)
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 5; Figure 2d, 2f and 2hHow it will be judged
The OFF linear plus exponential and double-exponential models (build_linear_slope_recovery_off_model, build_double_exponential_off_model, priors as in notebooks 21 and 24) fitted to the withdrawal-aligned OFF counts; ELPD as in figs8-null-model-worse. The Figure 2 legend says the improvement over the double exponential "was minimal", so no margin is required: reproduced when the linear plus exponential model has the highest ELPD of the OFF sigmoid models under either estimator. The difference and its standard error are reported; the paper's notebook 23 prints ELPD-LOO -6264.95 against -6267.36, difference 2.4, dse 3.5.
- fig2d-off-parameters
OFF-cell firing falls before withdrawal with sigmoid slope k = 2.852 s-1, then recovers with an exponential component of 4.760 s and a linear drift of 0.186 counts per second.
sponse slope of k = 2.852 s−1 (97% HDI: 2.333–3.392 s−1) ... component τ = 4.760 s (97% HDI: 3.711–5.861 s), together ... with a slower linear drift of klinear = 0.186 counts per sec
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 5; Supplementary Table S7How it will be judged
The OFF linear plus exponential fit of fig2g-switch-times; tau = 1/k2 (Table S7 caption). Reproduced when the three posterior means lie inside the paper's 97% HDIs; partially reproduced when two do. The paper gives k in s-1 here and in "total spikes per bin" in paragraph 6, and k3 in counts per second while the model's counts are per 50 ms bin summed over trials; the values are compared as the model's parameters, whatever their unit.
- tables9-s10-sigmoid-over-erf
A Gaussian CDF (erf) response for OFF cells fits significantly worse than the sigmoid, with either recovery model.
this fitted significantly worse than the sigmoid ... across all recovery models (see Supplemental Gaussian CDF
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 3; Supplementary Tables S9 and S10, Gaussian CDF Fits, Figure S9How it will be judged
The erf models of bayesian_cell_models.py (build_erf_off_model with double-exponential recovery, notebook 28; build_linear_erf_off_model with linear plus exponential recovery, notebook 29), with the Methods' priors (k_offset, k_scale ~ HalfNormal(10), k_slope ~ HalfNormal(4)), fitted to the withdrawal-aligned OFF counts; each compared by ELPD with the sigmoid model of the same recovery (notebooks 24 and 21). Readings of "significantly" as in figs8-null-model-worse. Reproduced when, for both recoveries, the sigmoid's ELPD is higher by more than one standard error of the difference (reading a) or two (reading b); partially reproduced when it is higher for both but by less. The paper's own notebook 29 prints ELPD-LOO -6263.78 for the linear erf model, above the -6264.95 of the linear sigmoid model, which disagrees with the text; this is recorded either way.
- figs1-departure-timing
Both populations leave baseline well before withdrawal; OFF firing has fallen 10% at 1.5 s before it and ON firing has risen 10% at 1.7 s before it, and both cross 50% of their change within a narrow window before withdrawal (OFF 726 ms, ON 630 ms before).
ready decreased by 10% from baseline at 1.5 s before the paw ... withdrawal (97% HDI: 1.746–1.306 s), and ON-cell firing in ... creased by 10% on a similar timescale, at 1.7 s before with ... a narrow pre-withdrawal window (OFF: 726 ms before; 97% ... HDI: 810–645 ms; ON: 630 ms before; 97% HDI: 690–570 ms)
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 6; Supplementary Table S2 and Figure S1How it will be judged
From the ON double-exponential and OFF linear plus exponential withdrawal-aligned fits, as calculate_rate_percentages_of_peak and calculate_rate_percentages_of_peak_OFF do (notebooks 18 and 21): for each posterior draw, the rate is predicted on a grid (ON 20000 points from -10 to 400 s; OFF 2000 points from -10 to 100 s); the baseline is the rate at the first grid point; the extremum is the grid maximum (ON) or minimum (OFF); the time of x% is the first grid time before the extremum at which the rate has moved x% of the way from baseline to extremum. Summary is the mean over draws and the 97% HDI. Reproduced when all four means lie inside the paper's HDIs; partially reproduced when two or three do.
- figs1-completion-asymmetry
ON firing reaches 90% of its peak 240 ms before withdrawal, while OFF suppression reaches 90% only 41 ms after it, so ON activation peaks before OFF suppression is complete.
(240 ms before, 97% HDI: 320–180 ms), while OFF-cell sup ... response (41 ms post, 97% HDI: 95 ms pre –180 ms post)
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraphs 6 and 7; Supplementary Table S2How it will be judged
As figs1-departure-timing, at 90%. The paper compares the two intervals and states no test (its HDIs, -0.32 to -0.18 s and -0.095 to 0.18 s, are disjoint). Reproduced when both means lie inside the paper's HDIs and the ON 97% HDI lies wholly before the OFF one; the fraction of paired draws, taken independently from the two fits, in which the ON time is earlier is reported.
- recovery-timing
After withdrawal ON cells return to 50% of their post-withdrawal baseline by 15.39 s but need 222 s to reach 90%; OFF cells return to 50% by 18 s and 90% by 83 s.
to 50% of their post-withdrawal baseline by 15.39 s after ... the withdrawal (97% HDI: 9.373-22.43 s), but required more ... reaching 90% of their baseline by 222 s (97% HDI: 183.6-248.8 ... s). OFF-cells increased back to 50% of their baseline by 18 s ... post-withdrawal (97% HDI: 15-21 s) but to 90% of their base ... line by 83 s (97% HDI: 82-84 s)
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 8; Supplementary Table S2How it will be judged
As figs1-departure-timing, after the extremum: the post-withdrawal baseline is the rate at the last grid point (ON 400 s, beyond the 100 s of data, as notebook 18 has it; OFF 100 s), and the time of x% recovery is the first grid time after the extremum at which the rate is within (100 - x)% of the extremum-to-baseline difference from baseline (the code's "fall to 10%" is the text's 90% recovery; Table S2 lists them so). Reproduced when all four means lie inside the paper's HDIs; partially reproduced when two or three do. As a variant, the ON times with the grid ending at 100 s.
- tables3-s7-withdrawal-steeper
Evoked transitions are sharper aligned to withdrawal than to heat onset; the OFF sigmoid slope is 2.852 withdrawal-aligned against 1.747 heat-aligned, and the ON rise time constant 0.653 s against 1.278 s.
creases was significantly steeper for the withdrawal-aligned ... data (2.852 total spikes per bin, 97% HDI: 2.333–3.392 spikes ... per bin), compared to the heat-aligned data (1.747 total spikes ... Similarly, ON-cells exhibited faster ... increases in firing when aligned to withdrawal ... nential time constants of 0.653 s (97% HDI: 0.570–0.739 s), ... compared to 1.278 s (97% HDI: 1.129–1.426 s) when aligned
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 9; Supplementary Tables S3, S5, S6 and S7How it will be judged
The ON double-exponential and OFF linear plus exponential models fitted to the heat-aligned counts of fig2-trial-counts (notebooks 18a and 21a, same priors as the withdrawal-aligned fits). The paper judges "significantly steeper" from the two HDIs (it states no test), so reproduced when the heat-aligned means lie inside the paper's HDIs and, for both cell types, the 97% HDIs of the withdrawal-aligned and heat-aligned estimates do not overlap (OFF k higher, ON tau_pre lower aligned to withdrawal). Partially reproduced when the HDIs are disjoint for one cell type only. The supplement notes r_hat problems for the heat-aligned ON fit; convergence diagnostics are reported.
- figs3-neutral-unmodulated
NEUTRAL cells (n = 32; 69 withdrawal-aligned and 70 heat-aligned trials) show no modulation linked to heat onset or withdrawal.
trast, NEUTRAL-cells (n = 32; 69 withdrawal-aligned and ... 70 heat-aligned trials) exhibited no discernible modulation ... associated with either heat onset or withdrawal behavior
Results, Evoked ON/OFF activity unfolds over multiple timescales, paragraph 9; Supplementary Figure S3How it will be judged
Trials of all NEUTRAL units of all files, cut and kept as in fig1c-psth-counts, under its two readings of the window start (Panel_1.ipynb uses the -30 to +150 s window and prints 69 and 70 trials). A claim of absence; the paper shows PSTHs and uses no test, so none is the paper's. The test run here is named as open: per trial, the spike count in the 2 s before the event against the 2 s after it (the window of the evoked change, Figure S1), Wilcoxon signed-rank over trials, two-sided, at p < 0.05, for each alignment. Reproduced when neither alignment shows a change and the counts are within tolerance; not reproduced when either does. The median change in rate and the same test for ON and OFF trials (as a positive control) are reported.
- ongoing-cell-counts
The ongoing-activity analyses use 25 ON, 25 OFF and 32 NEUTRAL cells recorded without external stimulation.
lyzed ON-, OFF-, and NEUTRAL-cells (NON = 25, NOFF = ... 25, NNEUTRAL = 32) recorded in the absence of external stim
Results, RVM ON- and OFF-cells exhibit slow quasi-periodic structure during ongoing activity, paragraph 1How it will be judged
Units of the "ongoing" and "evoked/ongoing" files (save_cell_activity.ipynb): ON and OFF from the ongoing files, NEUTRAL from both (12 + 20). Each ON and OFF cell (and each NEUTRAL cell of the ongoing files) contributes the 1500 s ending at the heat onset that closes its longest gap between heat onsets (Spike2Tools.get_event_windows, size 25 min; Methods, Data processing, "a 1500 s segment ... relative to the start of the first post–ongoing activity trial"); NEUTRAL cells of the evoked/ongoing files contribute their whole 960 s. Reproduced when the three counts are within tolerance and every ON and OFF cell has a 1500 s segment free of heat onsets; the length of each segment and of the buffer after the preceding trial (Methods, at least 230 s) are reported.
- fig3bc-psd-peaks
ON and OFF cells show clear low-frequency spectral peaks, often with harmonics, at a period of about five minutes (ON 286.86 s, FWHM 16.92 s; OFF 298.33 s, FWHM 16.12 s).
Both ON- and OFF-cells exhibited clear ... low-frequency spectral peaks, often accompanied by har ... istic period of approximately five minutes (TON ... peak = 286.86 ... s, full width at half maximum (FWHM) = 16.92 s; TOFF ... 298.33 s, FWHM = 16.12 s)
Results, RVM ON- and OFF-cells exhibit slow quasi-periodic structure during ongoing activity, paragraph 2; Figure 3b and 3cHow it will be judged
As Panel_3.ipynb and the supplement ("Spike train power spectra"): each cell's 1500 s segment (ongoing-cell-counts) binned at 5 ms, its periodogram with a Hann window over one segment, zero-padded (the code's FFT length is 50 times the spike count of the first ON cell, a resolution of 1.65e-4 Hz), DC removed, normalized to unit area (get_auto_correlation_freqs_with_rate); the class PSD is the mean over cells; the peak is the highest local maximum of the mean PSD between 0.0001 and 0.1 Hz (get_summary_psd_stats), and the period its inverse. The FWHM is computed per cell with scipy peak_widths at half height on the highest peak between 0.5 and 1.95 times the class peak frequency, and averaged over cells, in Hz. The paper reports the FWHM in seconds without saying how it was converted; the readings are (a) FWHM_Hz / f_peak squared, (b) 1/(f_peak - FWHM_Hz/2) - 1/(f_peak + FWHM_Hz/2), (c) the same as (b) from the half-height width of the mean PSD. The code's stored results (fwhm_results_ON.pkl, fwhm_results_OFF.pkl) give peaks at 288.9 s and 303.3 s and FWHMs of 2.13e-4 and 1.23e-4 Hz (about 17.8 and 11.4 s under reading b), which do not all match the text. The supplement ("Spike train power spectra", paragraph 3) describes a further estimate, which is also a reading: hierarchical bootstrapping at the rat level, N rats (the number in the class) resampled with replacement with all their cells, the mean PSD per iteration, its highest peak and that peak's FWHM taken per iteration, 1000 iterations; the period and FWHM are then the mean over iterations, with the FWHM in seconds under readings (a) and (b). Reproduced when both periods are within tolerance and the FWHMs match, under one reading; partially reproduced when only the periods do. Whether peaks near 600 s and 150 s (harmonics) are present is reported.
- fig3d-neutral-psd
NEUTRAL cells lack consistent low-frequency spectral peaks and instead show broad low-frequency components and prominent power above 10 Hz, reflecting regular tonic firing.
lacked consistent spectral peaks, instead showing broad low ... frequency components and prominent high-frequency power ... (> 10 Hz), reflecting regular, tonic firing
Results, RVM ON- and OFF-cells exhibit slow quasi-periodic structure during ongoing activity, paragraph 2; Figure 3dHow it will be judged
As fig3bc-psd-peaks, for the 32 NEUTRAL cells over their whole recordings (Panel_3.ipynb uses all NEUTRAL spike trains uncut; the supplement gives their lengths as 960 s and about 3000 s, zero-padded for a common grid). The paper does not define "consistent"; the reading used is the code's: the peak search of fig3bc-psd-peaks on the NEUTRAL mean PSD, and the per-cell peak frequencies between 0.5 and 1.95 times the ON peak frequency. Reproduced when the NEUTRAL mean PSD has no local maximum between 0.0025 and 0.0045 Hz (periods 220 to 400 s) that exceeds the median of the mean PSD over 0.001 to 0.01 Hz by more than its ON-cell counterpart's ratio divided by 2, and the NEUTRAL mean PSD has local maxima above 10 Hz that the ON and OFF mean PSDs lack. The bounds 220 to 400 s and the factor 2 are not in the paper and are named as open; the NEUTRAL peak the code finds (stored at 3.05e-4 Hz, beyond the window) and the spectra above 10 Hz are reported.
- fig3e-cv-isi
NEUTRAL cells fire regularly and fast (CV 0.35 ± 0.26, mean ISI 0.17 ± 0.25 s), while ON and OFF cells fire irregularly, with CVs well above 1 (ON 4.91 ± 4.32, OFF 3.70 ± 3.09) and broad ISI distributions (ON 1.25 ± 4.80 s, OFF 0.84 ± 1.36 s).
rapidly (Coefficient of Variation = 0.35 ± 0.26; mean Inter ... Spike Interval = 0.17 ± 0.25 s, ∼10 Hz). In contrast, ON- and ... with coefficients of variation well above 1 ( CVON = 4.91 ± ... 4.32, CVOFF = 3.70 ± 3.09) and broad interspike interval (ISI) ... distributions (ISION = 1.25 ± 4.80 s, ISIOFF = 0.84 ± 1.36
Results, RVM ON- and OFF-cells exhibit slow quasi-periodic structure during ongoing activity, paragraph 3; Figure 3e; Supplementary Table S1How it will be judged
As ongoing_CV_isi_table.ipynb: for each of the 82 cells of the ongoing and evoked/ongoing files, over its whole recording, the CV of its ISIs (elephant.statistics.cv, standard deviation over mean) and its mean ISI; then the mean and sample standard deviation over the cells of each class (Table S1). The tolerance is the paper's stated spread (protocol default), which is wide; reproduced when all six means are within it, the NEUTRAL mean CV is below 1 and both ON and OFF mean CVs above 1. Whether the six values also lie within 10% of the paper's is reported.
- fig4-nlml-minima
Scans of a periodic Gaussian process over the period show clear minima of the negative log marginal likelihood near 300 s for ON and OFF cells, with further minima at 600 and 900 s, while NEUTRAL cells, with one exception, have flat profiles without consistent minima.
ON- and OFF-cells showed clear minima in the ... negative log marginal likelihood (NLML) near 300 s, with ... additional subharmonic peaks at 600 and 900 s ... contrast, NEUTRAL-cells (with one exception) displayed flat
Results, A periodic Gaussian process reveals dominant slow oscillations in ON- and OFF-cell activity, paragraph 1; Figure 4How it will be judged
As GP_period_scoping.ipynb and the Methods (Gaussian process modeling, paragraph 1): each cell's rate from its spikes with a Gaussian kernel of 2 s sampled every 2 s (elephant.instantaneous_rate), edges removed, and a GPy GPRegression with a StdPeriodic kernel of fixed lengthscale 1.0 and variance 5.0 (GPy's default noise variance 1.0), its NLML evaluated at 200 periods evenly spaced. The Methods give 0.1 to 500 s; the Results and the code give 0.1 to 1500 s, which is used, since minima at 600 and 900 s cannot appear in a scan to 500 s (the 500 s scan is run as a variant for the 300 s minimum). The Methods remove the first and last 2 s; the code removes two samples (4 s), a variant. ON and OFF segments as in ongoing-cell-counts; NEUTRAL recordings whole. The paper does not define a "clear minimum"; the reading used is the code's display, each cell's NLML divided by its maximum: a cell has a minimum near p when its profile has a local minimum within 10% of p. Reproduced when a majority of ON and of OFF cells have a minimum near 300 s and the class means of the normalized profiles have local minima near 300, 600 and 900 s, and at most 4 of the 32 NEUTRAL cells (one, with 10% of 32 as tolerance) have a minimum near 300 s; partially reproduced when the ON and OFF part holds and more NEUTRAL cells have one. Counts per class are reported, and the stored scan PERIOD_SCOPE_RESULTS_200.npz is compared.
- fig5b-train-r2
A Poisson-type periodic Gaussian process fits the training data of all three cell classes with significant accuracy (pseudo-R2 different from 0; ON p = 0.0047, OFF p = 0.0004, NEUTRAL p < 0.0001).
All three cell classes achieved significant training accuracy ... (pON = 0.0047, pOFF = 0.0004, pNEUTRAL < 0.0001)
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 2; Figure 5b; Supplementary Figure S4How it will be judged
The GP fit, from the Methods (Gaussian process modeling, paragraph 2) and Poisson_GP_main_fit_notebook.ipynb with non_poisson_gpytorch_model.py: spikes in 5 s bins; each cell trained on all but its last 300 s (ON and OFF 1500 s segments split at 1200 s; evoked/ongoing NEUTRAL cells, 960 s, at 660 s as the notebook has it, with params.json's 700 s as a variant; ongoing NEUTRAL cells over their whole recording, the Methods' "full recording period" (Data processing, paragraph 2), less its last 300 s, with the code's split at 1200 s of a 1500 s segment as a variant); a sparse variational GP (GPyTorch ApproximateGP, fixed inducing points evenly spaced over the training span; the Methods give 200 inducing points, the code 100, a variant), constant mean, scaled periodic kernel, lengthscale prior N(0.1, 0.1) on the raw lengthscale, period initialized at 300 s and lengthscale at 0.1, mean and outputscale initialized by set_mean_and_scale; rate 10 softplus(x); generalized Poisson likelihood with theta = mu(1 - alpha), alpha < 1, prior N(0, 0.5); uniform contamination on 0 to 200 counts with p_noise 0.001 (Methods; the code fixes 0.01, a variant); Adam, learning rate 0.05, for 10000 iterations (params.json; the notebook's last call uses 100, a variant); seeds 0 (torch and pyro). Pseudo-R2, which the paper does not define, is the code's (compare_log_liks): 1 minus the summed expected log likelihood of the periodic GP over that of a constant-mean GP with a diagonal kernel and the same likelihood, on the training and the test bins. The test, as bootstrap_R2_across_animals.ipynb: the observed median pseudo-R2 of a class against the medians of 10000 permutations that flip the sign of all cells of one animal together (the Results, paragraph 2, "all cells per animal"; animals are dandiset subjects); the code groups by recording file, run as a variant, which is the same when each file is one animal. p is the fraction with absolute null median at least the absolute observed median. Reproduced when all three classes have p < 0.05 with positive medians; the median per class is reported, and the paper's stored values (combined_data.csv) are compared.
- fig5b-train-no-difference
Training pseudo-R2 does not differ between the three classes (permuted, Bonferroni-corrected Mann-Whitney U tests; Kruskal-Wallis p = 0.117).
paring R2 values using a similarly permuted and Bonferroni ... corrected Mann Whitney U test revealed no significant differ ... ences between training R2 for the three groups
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraphs 2 and 3; Supplementary Figures S6 and S7How it will be judged
A claim of absence, judged on the paper's tests as bootstrap_R2_across_animals.ipynb runs them on the training pseudo-R2 of fig5b-train-r2: for each pair of classes, the Mann-Whitney U statistic against 1000 permutations that relabel the cell classes of each recording file as a block (mannwhitney_blockwise, seed 42; Figures S6 and S7 say the permutation is within animals, so blocks by subject are run as a variant), two-sided by distance from the null median, Bonferroni-corrected over the three pairs; and the Kruskal-Wallis H against 1000 such permutations over the three labels (kruskal_blockwise, seed 42). Reproduced when no corrected pair p and not the Kruskal-Wallis p is below 0.05; not reproduced when any is. The statistics and p-values are reported either way.
- fig5b-test-r2
On held-out data ON cells generalize significantly (p = 0.0067), with most cells above 0; OFF cells as a population do not (p = 0.3127), mixing well and poorly predicted cells; NEUTRAL cells fall significantly below baseline (p = 0.0033).
significant generalization to unseen data (p = 0.0067), with ... OFF-cells, as a population, did not show reliable ... predictive structure (p = 0.3127) ... cells performed significantly below baseline (p = 0.0033)
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 3; Figure 5bHow it will be judged
The test of fig5b-train-r2 on the test pseudo-R2. Reproduced when ON has p < 0.05 with a positive median and more than half of ON cells above 0, OFF has p >= 0.05, and NEUTRAL has p < 0.05 with a negative median; partially reproduced when two of the three hold. The OFF part is a claim of absence and is judged on this test alone.
- fig5b-test-group-differences
Test pseudo-R2 of NEUTRAL cells differs from that of ON cells (p < 0.001) and OFF cells (p = 0.006), ON and OFF cells do not differ (p = 0.939), and the Kruskal-Wallis test across the three is significant at test time (p < 0.001).
vealed significant differences between NEUTRAL-cells and ... both ON- and OFF-cells (p < 0.001 and p = 0.006, respec ... tively), while ON- and OFF-cells did not differ significantly ... at test (p = 0.939). Kruskal–Wallis tests confirmed signifi ... cant group differences at test time (p < 0.001)
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 3; Supplementary Figures S6 and S7How it will be judged
The tests of fig5b-train-no-difference on the test pseudo-R2. Reproduced when both NEUTRAL pairs and the Kruskal-Wallis test have p < 0.05 and the ON-OFF pair has p >= 0.05 (a claim of absence, judged on this test); partially reproduced when the Kruskal-Wallis test and one NEUTRAL pair hold. The other permutation scheme in the notebook (clustered_mannwhitney_u, shuffling labels within each file, which printed p = 1.0 for both NEUTRAL pairs) is shown as a variant.
- table1-hyperparameters
The fitted GP periods of ON and OFF cells cluster near 300 s (ON 297 ± 22 s, OFF 304 ± 27 s); NEUTRAL cells give 322 ± 54 s, which the paper reads as a weak constraint given their low latent variance and poor prediction.
periods for both ON- and OFF-cells clustered near 300 s, ... corroborating the frequency-domain analyses
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 4; Table 1; Figure 5cHow it will be judged
The fitted kernel periods and lengthscales of fig5b-train-r2; mean and sample standard deviation over cells of each class (Panel_4.ipynb, make_latex_table). The tolerance is the stated spread over cells. Reproduced when all six means are within tolerance; partially reproduced when the three periods are. Since the periods start at 300 s, the change from the initial value per cell is reported.
- fig5c-dispersion
ON and OFF cells have positive dispersion parameters (ON 0.63 ± 0.25, p = 0.0011; OFF 0.36 ± 0.53, p = 0.0105), consistent with bursty over-dispersed firing, while NEUTRAL cells have negative dispersion (-0.9 ± 0.63, p < 0.001), firing more regularly than Poisson.
ON- and OFF-cells exhibited positive dispersion parameters ... (α > 0; pON = 0.0011, pOFF = 0.0105), consistent with ... showed negative dispersion (pNEUTRAL < 0.001)
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 5; Table 1; Figure 5c; Supplementary Figure S5How it will be judged
The fitted dispersion alpha of fig5b-train-r2 per cell; the sign-flip permutation test of fig5b-train-r2 on alpha (10000 permutations, Figure S5; the notebook seeds 1). Reproduced when ON and OFF have p < 0.05 with positive medians, NEUTRAL has p < 0.05 with a negative median, and the three means are within the stated spread.
- table1-latent-variance-order
The latent GP variance separates the classes; ON cells have the largest modulation amplitude (outputscale 17 ± 26), OFF cells intermediate (3.1 ± 4) and NEUTRAL cells a near-flat latent structure (0.1 ± 0.2).
cells showed the largest modulation amplitude, OFF-cells ... intermediate variance, and NEUTRAL-cells near-flat latent
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 5; Table 1How it will be judged
The fitted kernel outputscale of fig5b-train-r2 per cell. The paper states the ordering of the class means and no test. Reproduced when the class means order ON > OFF > NEUTRAL and are within the stated spread; the class medians, whose order may differ given the spread, are reported.
- fig5d-phase-profiles
Single-cycle phase-normalized reconstructions of the GP fits show cell-specific oscillatory signatures in ON and OFF cells, while NEUTRAL cells have near-constant phase profiles.
revealed cell-specific oscillatory signatures in ON ... and OFF-cells, whereas NEUTRAL-cells showed near ... constant phase profiles
Results, A Poisson Gaussian process predicts slow quasi-periodic fluctuations in ongoing RVM firing, paragraph 6; Figure 5dHow it will be judged
The Figure 5 legend says the panel shows the GP rates "phase aligned" and "adjusted for found period length"; the paper gives no further method and no test, and the code for the panel was not checked here. Reading, named as open: for each cell, the predicted rate of its GP fit of fig5b-train-r2 over one fitted period, on a phase grid from 0 to 1 with zero phase at the fit's maximum (as get_phase of figs2-phase-latency); the modulation depth is (maximum minus minimum) over the mean of that cycle. Reproduced when the median depth of NEUTRAL cells is below the medians of both ON and OFF cells; the per-cell depths and profiles are reported. The comparison and the depth measure are not in the paper.
- fig6c-coherence-counts
A subset of cells of every class shows significant low-frequency coherence with heart rate (5 of 32 NEUTRAL, 4 of 25 OFF, 8 of 25 ON).
A subset of cells (5/32 NEUTRAL, 4/25 OFF, 8/25 ON) ... showed significant low-frequency coherence with heart rate
Results, Low-frequency coherence between RVM activity and heart rate, paragraph 3; Figure 6b and 6cHow it will be judged
As the Methods (Coherence estimation) and 07_mt_sig_test_phase_shifts.ipynb: for each cell, its spikes and the EKG beats of its recording over the cell's ongoing segment (ON, OFF and ongoing NEUTRAL 1500 s segments of ongoing-cell-counts; evoked/ongoing NEUTRAL whole 960 s; the Methods say only "the recording", so each cell's whole recording is a second reading), each binned at 2 s; 500 s segments with 50% overlap, each demeaned; multitaper spectra with DPSS tapers, NW 3.5, K 6, and the magnitude-squared coherence averaged over segments; 1000 surrogates with the heart-rate segments phase-randomized (seed 1); the threshold is the 99.9th percentile of the surrogates' maximum coherence over all frequencies (max-statistic, alpha 0.001). A cell is coherent when its coherence exceeds the threshold at any frequency, as the code counts (the code does not restrict the frequency; the Results say "low-frequency"; the count restricted to below 0.01 Hz is shown as a variant). Tolerance 10% of each count, so the counts must match exactly; reproduced when all three do under one reading, partially reproduced when each is within 2 of the paper's. Because the surrogates are random, the counts at three other seeds are reported.
- fig6d-heart-rate-peaks
Peak coherence frequencies do not match the ~300 s period exactly, but the heart-rate power spectrum has peaks at multiples of this period.
spectral peaks in heart rate were observed at ... multiples of this period (Figure 6d)
Results, Low-frequency coherence between RVM activity and heart rate, paragraph 3; Figure 6dHow it will be judged
The heart-rate spectrum is described in the supplement ("Spike train power spectra", last paragraph: computed by bootstrapping "in the same manner as for the cells", one heart rate per animal) and the Figure 6 legend ("Combined averaged heart rate power spectrum over all animals"); no code for it was found in the repository, and the paper does not say whether the EKG beats are treated as a point process or converted to a rate, which segments are used, or what "multiples" means (periods of 600 and 900 s, or frequencies of 2 and 3 times 1/300 Hz). Readings: the EKG beat train of each ongoing and evoked/ongoing recording (one per animal), (a) as a spike train over the same 1500 s segment and (b) over the whole recording, with the PSD method of fig3bc-psd-peaks; the mean over animals; a "peak at a multiple" is a local maximum within 10% of 600 or 900 s (subharmonics) or of 150 or 100 s (harmonics). Reproduced when, under one reading, at least one such peak is the highest peak below 0.01 Hz; the lowest-frequency peak (marked in Figure 6d) is reported. The rule "highest peak below 0.01 Hz" is not in the paper and is named as open. The first part of the statement is reported, not judged: the frequency of peak coherence of each coherent cell of fig6c-coherence-counts, against 1/300 Hz.
- figs2-phase-latency
Withdrawal latency shows no systematic relation to the predicted firing rate or oscillatory phase of the ON and OFF GP fits at stimulus onset.
As shown in Supplementary Figure S2d,e,f, there was no systematic relationship between normalized withdrawal latency and either predicted firing rate or oscillatory phase at stimulus onset.
Supplementary Information, Ongoing activity phase did not affect withdrawal latency, paragraph 4; Supplementary Figure S2d to S2f; Discussion, paragraph 4How it will be judged
As phase_analysis.ipynb: for each ON and OFF cell of the ongoing files, its GP fit of fig5b-train-r2 extrapolated to the heat onsets of its whole recording; the predicted count per bin normalized between the minimum and maximum of that fit; the phase from the fitted period with zero phase at the fit's maximum (get_phase); latencies (withdrawal minus heat onset) z-scored per cell. A claim of absence, but the paper only inspects the plots and states no test, so the test is named as open: Spearman correlation of normalized latency with normalized predicted rate, and a circular-linear correlation of latency with phase, each for ON and OFF trials pooled within class, two-sided at p < 0.05. Reproduced when none of the four is significant; not reproduced when any is. The correlations and the number of trials are reported.
- methods-withdrawal-latency
The heat ramp was ended at paw withdrawal, with mean withdrawal latency 8.7 ± 1.1 s (mean ± SD) and withdrawal temperature 49.0 ± 2.0 C.
temperature: 49.0 ± 2.0 °C, latency: 8.7 ± 1.1 s)
Methods, Experiments, paragraph 2How it will be judged
Latency is the time from each heat onset to the next withdrawal, when the withdrawal comes before the next heat onset (the pairing of Spike2Tools.get_tail_to_heat_lengths). The paper does not say over which trials the mean is taken; the readings are (a) all paired trials of all files, (b) the mean of per-animal means. Reproduced when, under one reading, the mean is within the stated SD. The temperature is tested only if the dataset holds it; otherwise that part is reported as not tested.
- methods-heart-rate
Heart rate, from the EKG, was 391.1 ± 34.4 bpm (mean ± SD across animals).
was 391.1 ± 34.4 bpm, with mean range over the protocol
Methods, Experiments, paragraph 1How it will be judged
Heart rate from the EKG beat times of each recording as 60 over the mean inter-beat interval. The paper does not fix the averaging; the readings are (a) per recording, then mean and SD over recordings, (b) per animal (recordings of one subject pooled), then over animals. Reproduced when, under one reading, the mean is within the stated SD; the SD is reported.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none