Cortical preparatory activity: representation of movement or first cog in a dynamical machine? (DANDI:000140)

Claims

From the paper Cortical preparatory activity: representation of movement or first cog in a dynamical machine?

Reproducing Churchland et al. (2010) from dandiset 000140 16

Notebook reproduce_churchland_2010.ipynb

Reproduced5

ReproducedResults, Correlations between preparatory and peri-movement tuning, paragraph 2; Figure 4Erests on an assumption
When peri-movement activity is taken at single time points, its correlation with preparatory tuning starts to fall about 200 ms before movement onset and is near zero by movement onset.
The correlation with preparatory activity began to decline ... prior to movement onset, and reached values near zero by movement onset
Results, Correlations between preparatory and peri-movement tuning, paragraph 2; Figure 4E
Under all 18 O1 range readings the mean correlation at movement onset is below 0.1 and below half its value at -200 ms, and the largest 50 ms drop starts 270 to 210 ms before onset. Main reading (fig4 units, 19) 0.295 at -200 ms and -0.078 at onset; all units with preparatory tuning (93) 0.62 at -400 ms, 0.39 at -200 ms, 0.12 at -100 ms and -0.00 at onset. Trial noise also meets the criterion, so the pass carries little evidence. With condition labels shuffled it is met in 18 of 20 shuffles for the fig4 units and 20 of 20 for all units with > 5 spikes/s (r at -200 ms 0.28 to 0.37, at onset -0.028 to 0.035), because the preparatory and movement-locked means share the noise of the same 1 to 4 trials and the preparatory window ends about 220 ms before onset. With the two means from disjoint trials the correlation at -200 ms is only 0.12 to 0.18 for all units, and the criterion is met in 4 of 12 readings (fig4 units) and 9 of 12 (all units). Under O1 SD (3 to 8 units) the value at onset is 0.091 to 0.244, partially reproduced in 11 of 12 readings; in the twelfth (O3a, O4a, O2 ii, 5 units) r at onset is 0.091 but the largest drop starts at +90 ms, so it is neither a pass nor a partial.
How the test differed from the paper’s method
- Assumed O1 is the range across conditions (see fig4-broad-near-zero).
- The time axis uses go-locked means from -200 to +200 ms and movement-locked means from -400 to +400 ms, in 10 ms steps (Supplemental Experimental Procedures, the spliced trace); the judgement uses the movement-locked part.
ReproducedResults, Correlations between preparatory and peri-movement tuning, paragraph 3; Supplemental Figure 1
Population-average PSTHs grouped by each neuron's preparatory preference show weak peri-movement differences that shrink with time, with little relationship to preparatory tuning by movement onset.
In population averages ( Supp. Fig. 1 ) the lack of correlation averages out and the means are dominated by the weak agreement between preparatory and peri-movement tuning. However, such effects grow dramatically smaller with time. By movement onset even population averages show little relationship with preparatory tuning
Results, Correlations between preparatory and peri-movement tuning, paragraph 3; Supplemental Figure 1
In all 36 readings, normalised and not, the spread between the strongest and weakest preparatory conditions at movement onset is below half the preparatory spread (ratio 0.186 in the main normalised reading; -0.175 to 0.295 over readings), with the strongest trace on top at -150 ms. Trial noise also meets the criterion. With condition labels shuffled it is met in 11 of 20 shuffles, and with the preparatory and movement-locked means from disjoint trials in 2 of 12 readings. This suggests that the strongest trace being on top at -150 ms comes mostly from noise shared within the same trials, but does not show it, as each disjoint half uses about half the trials and the added noise could also hide a real effect. The pass carries little evidence.
How the test differed from the paper’s method
- Each unit is normalised by its maximum condition-mean rate over the go-locked and movement-locked spans stored (-300 to +200 ms and -400 to +900 ms); the legend says "normalizing each neuron's activity by its maximum rate".
ReproducedResults, Controls, paragraph 1; Figure 7A and legend
A control peri-movement space in which each condition's peri-movement response is inverted at random on half the conditions performs poorly.
The PCA-based 'shuffled-space' performed poorly
Results, Controls, paragraph 1; Figure 7A and legend
In all 24 readings (six O3/O4 settings, O1 range and SD, the coin per unit and condition or per condition) the mean best performance over 20 shuffles is below half the intact best and below the best kinematic performance. Main reading -0.097 against -0.044 intact and -0.053 kinematic. Under O1 range the intact space is itself below zero, so below half the intact best is met trivially. The evidence is the O1 SD readings (7 or 8 units), where the shuffled mean is -0.065 to 0.041 against an intact 0.129 to 0.249.
ReproducedResults, Comparison of anterior versus posterior sites, paragraph 1rests on an assumption
For J-array, the preparatory to peri-movement tuning correlation does not depend on anterior-posterior recording location (r = 0.20, p = 0.27).
Regressing tuning correlation (as in Figure 4 ) against AP location yielded little or no effect ... and r =0.20, p =0.27 for the four datasets
Results, Comparison of anterior versus posterior sites, paragraph 1
What the paper reports
- J-array, tuning correlation regressed on AP location: linear regression, r = 0.20, p = 0.27
In all 17 readings with units on both arrays the tuning correlation does not differ by array (p 0.538 to 1.000). Main reading r = 0.105, p = 0.668, 10 PMd and 9 M1 units (mean r -0.060 and -0.099). The outcome holds under either mapping of the arrays; swapping them changes only the sign of r. With 19 units and two locations the test has little power; the paper's r = 0.20 would not be detected here either.
How the test differed from the paper’s method
- The electrodes have no positions, so AP location is the array, PMd anterior and M1 posterior (Supplemental Figure 3 legend). Regression on a 0/1 location, the two-group comparison the criterion names as the variant for array identity.
- Assumed the array of a unit is the first digit of its id, 1 for PMd and 2 for M1, as units with prefix 1 map onto the PMd electrodes of the table; every unit's electrodes entry points to the PMd group. Swapping the arrays changes only the sign of r.
ReproducedResults, Tuning from a dynamical systems perspective, paragraphs 2 and 3; Figure 8 and legend
In a simulated two-unit rotational linear system with 25 random initial states, the single-unit correlation between initial (preparatory) and later (peri-movement) activity is near zero (0.08), yet a PD in the PCA-based peri-movement space captures the initial state perfectly.
The correlation of preparatory activity with peri-movement activity (averaged over the gray interval) was 0.08 ... The PD points towards the pattern recruited by the activity of unit one (panel A ), and perfectly captures preparatory
Results, Tuning from a dynamical systems perspective, paragraphs 2 and 3; Figure 8 and legend
Over 1000 simulations the median correlation between unit 1's initial rate and its mean over the 360-step released interval is 0.0044 (initial states uniform in the unit disk) and 0.0115 (in a square), within 0.08 +/- 0.3; the median |r| is 0.111 and 0.147, below 0.3; the PD in the 2-D PCA space of the released trajectories gives leave-one-out performance of 1.0 in every simulation.

Partially reproduced1

Partially reproducedResults, PD performance in a variety of spaces, paragraphs 1 to 3; Figure 6Dverifier disagreed at firstrests on an assumption
For the J-array dataset, the cross-validated performance of a preferred direction in capturing preparatory tuning is highest when the PD is expressed in the PCA-based peri-movement space, compared with the kinematic space and the task spaces.
For every dataset, the highest level of performance was achieved ... when PDs were expressed in the peri-movement space
Results, PD performance in a variety of spaces, paragraphs 1 to 3; Figure 6D
Main reading (O1 range, O3a, O4a; 93 units) best peri-movement performance -0.044 at k = 2, best kinematic -0.053, best task space (initial velocity) -0.048. Of 96 readings 64 meet the criterion and 32 are partial (initial velocity or kinematic at or above the peri-movement space), the partial ones under O1 range with O3a and O4b, O3b and O4a, or O3b and O4b, and under O1 SD with O3b and O4b. The criterion asks that it hold under each reading, so the outcome is partial. The differences are small against their spread. In no reading is the peri-movement space better than the best other space at p < 0.05 (paired t p 0.47 to 0.98; main difference 0.004, SE 0.018). With the T matrix built from the included units only, O1 range gives the same (reproduced or partial in each of 6 settings), while under O1 SD the peri-movement space falls to -0.022 to 0.101 and 6 to 8 spaces match or beat it. Under O1 range all spaces perform below zero, worse than predicting the mean rate, but above a condition-shuffled null (peri-movement -0.214 to -0.092, kinematic -0.138 to -0.108). Under O1 SD (8 units, T of all units) the peri-movement space is best with positive performance, 0.209 against 0.136 kinematic and 0.133 task. Repeating the kinematic variables (reading a) leaves performance unchanged (largest difference 8e-15).
How the test differed from the paper’s method
- Assumed O1 range (see fig4-broad-near-zero). Under O1 SD with the T matrix of the included units the claim is not reproduced; under O1 range every reading gives reproduced or partial.
- The units of the peri-movement T matrix are read two ways, all recorded units (main) and the units passing the inclusion rule. The Methods say each row "contained data from all neurons except the neuron whose preparatory activity we were trying to capture" and that "Analysis of preparatory responses was restricted to neurons with at least 10 spikes/s of preparatory tuning".
- The upper limit uses the preparatory matrix of all units except the fitted one, as the peri-movement space does (Methods, PCA-based spaces); the supplement says "all neurons".
- Task spaces follow the Methods' list (Task spaces). Details they leave open, set here and stated in maze_prep.py, are the end of the reach (last time within 800 ms of onset with speed at least 10% of peak), the peak velocity components (the signed value of largest magnitude), and the segments (equal path length, with equal duration as a variant, which gives the same outcome).
- Acceleration for the kinematic space is the derivative of the dataset's hand velocity after a 20 ms Gaussian, a standard choice.

Not reproduced8

Not reproducedResults, Correlations between preparatory and peri-movement tuning, paragraph 1; Figure 4D
For the J-array dataset, the correlation across conditions between a unit's preparatory and peri-movement firing rates is positive on average but small (mean Pearson r = 0.13, p = 0.03; mean Spearman r = 0.11).
Across the four datasets the mean correlation was positive (p=0.01, p=0.01, p=0.12, p=0.03) but only very modestly so (mean r = 0.21, 0.19, 0.06 and 0.13). Non-parametric correlations were similarly modest (mean Spearman's r = 0.20, 0.20, 0.06 and 0.11)
Results, Correlations between preparatory and peri-movement tuning, paragraph 1; Figure 4D
What the paper reports
- mean Pearson correlation against zero, J-array: not named, p = 0.03 (n = J-array units passing the preparatory and peri-movement tuning criteria (count not given))
Under none of the 18 readings with units (O1 x O2 x O3 x O4) is the mean correlation within 0.13 +/- 0.013 or positive at p < 0.05. Mean Pearson r ranges from -0.148 to 0.095 and mean Spearman r from -0.061 to 0.061; the one reading with p < 0.05 (O3a, O4b, O1 range, O2 i, 13 units, r = -0.148) is negative. The main reading (O1 range, O2 i, O3a, O4a) gives 19 units over 27 conditions, mean r -0.079 (t-test p = 0.086) and Spearman -0.020; with O2 ii 88 units and mean r 0.011 (p = 0.662). With condition labels shuffled across trials (20 shuffles, no tuning) the mean r is -0.106 to 0.074, so the main value is within what trial noise gives. Under O1 SD no unit passes with O2 i and 3 to 5 with O2 ii. The condition means rest on 1 to 4 trials (66 to 78 trials kept), and the split-half reliability of preparatory tuning has a median of 0.20 over units, so a mean correlation of 0.13 could not be resolved here; the outcome describes these 100 trials, not the paper's 2166 trials per neuron.
How the test differed from the paper’s method
- 27 conditions with 1 to 4 trials each after the delay rule, against 108 conditions and about 20 trials each in the paper's J-array dataset.
- Spike trains smoothed with a Gaussian of SD 20 ms before averaging (Supplemental Experimental Procedures, Computing the mean firing rate versus time; the width is read as the SD, a standard choice).
Not reproducedResults, Correlations between preparatory and peri-movement tuning, paragraphs 3 and 4; Figure 4A to 4Dverifier disagreed at firstrests on an assumption
Correlations between preparatory and peri-movement tuning span a broad range centred near zero, with no separate populations of concordant (r near 1) and opposing (r near -1) tuning; most neurons have weakly correlated tuning.
different neurons show markedly different relationships between preparatory and peri-movement activity, with a broad range of correlations centered near zero ... Instead, most neurons have weakly correlated tuning
Results, Correlations between preparatory and peri-movement tuning, paragraphs 3 and 4; Figure 4A to 4D
Under O1 range (12 readings) the median r is near zero (-0.098 to 0.016), 92% to 100% of units have |r| < 0.5, at most 1.2% are above 0.8 and none below -0.8, but the interquartile range is 0.173 to 0.286, below the 0.3 the criterion requires, in every O1 range reading. The criterion has no partial rule, and the condition that fails is the one that tests the breadth the claim states. The conditions that hold are also met by noise. With condition labels shuffled (20 shuffles) the IQR is 0.069 to 0.422 and the mean r -0.106 to 0.074. Under O1 SD (3 to 5 units) all 6 readings pass.
How the test differed from the paper’s method
- Assumed O1 is the range of the preparatory rate across conditions, as the Methods describe the inclusion rule as a rate difference ("Firing rate modulation was typically lower ... >5 spikes/s of preparatory tuning"); under the SD reading the claim is reproduced, on 3 to 5 units.
Not reproducedResults, PD performance in a variety of spaces, paragraph 3; Figure 6A to 6D
Taken one dimensionality at a time, performance is almost always higher in the peri-movement space than in other spaces of the same dimensionality, the exception being the 3-D targets + instructed speed and endpoints + maximum speed spaces for the speed task.
Even if we consider each dimensionality individually, performance was almost always higher for the peri-movement space. An exception was that, among 3-dimensional spaces, the 'targs+instr. spd.' and 'endpoints + max. spd.' performed very well for the datasets collected using the speed task
Results, PD performance in a variety of spaces, paragraph 3; Figure 6A to 6D
The peri-movement space wins 7% to 36% of the same-dimension comparisons over the 96 readings of fig6d, never the 80% the criterion requires. Its performance falls steeply with k (main reading -0.104 at k = 3, -1.369 at k = 6) while the kinematic space falls slowly (-0.081 and -0.191).
Not reproducedResults, PD performance in a variety of spaces, paragraph 3; Figure 6C and 6Drests on an assumption
In the maze task the endpoints + maximum speed space, which did well for the speed task, does not perform well.
The former is undefined for the maze task, but the latter did not perform well
Results, PD performance in a variety of spaces, paragraph 3; Figure 6C and 6D
Under O1 range (12 readings) the endpoints + max. speed space (-0.126 to -0.147) is below the 3-D peri-movement space by 0.011 to 0.065, less than the 0.1 required (main reading -0.143 against -0.104, SE of the per-unit difference 0.025). Under O1 SD all 12 readings pass with the T matrix of all recorded units (-0.040 to 0.018 against 0.101 to 0.209); with the T matrix of the 7 or 8 included units the 3-D peri-movement space gives -0.088 to 0.045 and none of the 6 settings passes.
How the test differed from the paper’s method
- Assumed O1 is the range across conditions (see fig4-broad-near-zero); under the SD reading with the T matrix of all recorded units the claim is reproduced.
Not reproducedResults, PD performance in a variety of spaces, paragraph 2; Figure 6A to 6Drests on an assumption
Performance of the PCA-based spaces typically improves with dimensionality and then declines above about 10 dimensions, an indication of overfitting.
Performance typically improved with dimensionality ... Performance typically declined at dimensionalities >10, an indication of over-fitting
Results, PD performance in a variety of spaces, paragraph 2; Figure 6A to 6D
Under O1 range the best k is 2 in all 6 readings, so performance does not rise with dimensionality; the decline above 10 holds (main reading -4.069 at k = 10, -44.316 at k = 20). Under O1 SD the rise and the decline both hold in 3 of 6 readings (best k = 4), and in the other 3 the best k is 2.
How the test differed from the paper’s method
- Assumed O1 is the range across conditions (see fig4-broad-near-zero).
Not reproducedSupplemental Experimental Procedures, Estimating the upper limit of PD performance; Results, PD performance in a variety of spaces, paragraph 2
Preparatory activity is moderately high dimensional; depending on the dataset 7 to 10 dimensions capture 90% of its variance.
depending on the dataset, 7-10 dimensions were required to capture 90% of the data variance
Supplemental Experimental Procedures, Estimating the upper limit of PD performance; Results, PD performance in a variety of spaces, paragraph 2
90% of the variance of the condition-mean preparatory matrix needs 16 or 17 components under O1 range (91 to 93 units) and 5 or 6 under O1 SD (7 or 8 units); none of the 12 readings is within 6.3 to 11, z-scored or not. With the condition labels shuffled across trials the count is 16 to 17, the same as with the true labels, so under O1 range the count measures trial noise. The count is bounded here by 26.
Not reproducedResults, Controls, paragraph 1; Figure 7Brests on an assumption
Sliding the start of the peri-movement window later costs little; with the window starting at movement onset, performance is still 97% (monkey J) and 94% (monkeys A, B) of its original value.
Even when the window began at movement onset ... performance was still at 97% (monkey J), and 94% (monkeys A, B) of its original value
Results, Controls, paragraph 1; Figure 7B
Under O1 SD the ratio of performance with the window starting at movement onset to that at -150 ms is 0.471 to 0.761 in the 6 O3/O4 settings, outside 0.97 +/- 0.097 (main settings 0.209 to 0.105, ratio 0.501). Under O1 range performance at -150 ms is below zero (-0.046 to -0.023), so the ratio is undefined; it falls from there to -0.059 to -0.046 at onset. Only the monkey J value under reading (a) can be computed; the monkeys A and B value (0.94) is not in the dandiset.
How the test differed from the paper’s method
- Assumed O1 SD (O1b). Under the O1 range reading adopted elsewhere (see fig4-broad-near-zero) performance at -150 ms is below zero and the ratio is undefined, so the outcome rests on the O1b readings (7 or 8 units) alone.
- Start times -150 to +100 ms in 50 ms steps, end fixed at +800 ms, best k up to 14 for each start (Figure 7B legend, "the best dimensionality for that dataset/start-time").
Not reproducedResults, Stability of the PD, paragraphs 1 and 2; Figure 7D and legend
The PD measured on two random sets of 25 conditions is more stable (higher dot product) in the peri-movement space than in the task, kinematic and EMG spaces, for every comparison and every dataset.
Yet for every comparison, and for all four datasets, the PD was most stable when measured in the peri-movement space
Results, Stability of the PD, paragraphs 1 and 2; Figure 7D and legend
Over the six O3/O4 settings, both O1 readings and raw or z-scored axes (72 comparisons) the peri-movement space is more consistent in 46. It is more consistent than endpoints + max. speed in 24 of 24 (main reading 0.617 against 0.523) and than the kinematic space at its best k in 21 of 24 (main, k = 1, 0.587 against 0.563), but than segment directions at k = 6 in only 1 of 24 (main 0.416 against 0.525; O1 SD with O3a and O4a 0.545 against 0.615).
How the test differed from the paper’s method
- The paper does not give the scale of task-space axes, which sets a PD's direction when units are mixed (mm and mm/s); raw and z-scored axes are both run and agree on the segment-directions comparison in 11 of 12 pairs (they differ under O1 SD with O3a and O4b).
- Draws in which the unit's rate is the same in all drawn conditions (a zero PD) are dropped; 100 repeats per unit, the claim's standard choice.

Blocked2

BlockedResults, Comparison of anterior versus posterior sites, paragraph 1; Figure 7C and legend· Otherverifier disagreed at first
The peri-movement space captures preparatory tuning at both anterior and posterior sites, with normalised performance slightly better at anterior sites.
The ability of the peri-movement space to capture tuning also held for both anterior and posterior sites ( Figure 7C ) ... Performance was slightly better at anterior sites
Results, Comparison of anterior versus posterior sites, paragraph 1; Figure 7C and legend
The normalised measure the criterion uses does not exist here. Under O1 range the upper-limit performance is at or below zero for every subset (-0.019 to -0.097), so performance cannot be normalised by it. Raw 6-D peri-movement performance is far below zero at both arrays, -1.535 (prefix 1, 44 units) and -1.087 (prefix 2, 49 units) with the space from the half's units, and -1.454 and -1.293 from all units, against -1.008 and -1.400 for random halves. Under O1 SD (4 and 4 units) the upper limits are positive (0.145 and 0.184) but the 6-D performance is not (normalised -12.048 and -2.915, random halves -7.282), so a threshold at 70% of the random-half value has no meaning either. Under O1 SD the upper limit of all 8 units (0.123) is also below the best peri-movement performance (0.209), so it is not a ceiling. Which half is anterior is also not known (prep-activity-more-anterior); with prefix 1 as PMd the anterior half is lower in raw performance, with the arrays swapped higher.
How the test differed from the paper’s method
- AP halves are the two arrays (see ap-location-no-effect-j-array), not a median split of positions.
- O1 range, O3a and O4a (see fig6d-perimovement-best-j-array); 20 random halves of the size of the prefix-1 array.
BlockedResults, Comparison of anterior versus posterior sites, paragraph 1· Metadata insufficientverifier disagreed at first
Preparatory activity is more prevalent at anterior sites.
We similarly found preparatory activity to be more prevalent at anterior sites (data not shown)
Results, Comparison of anterior versus posterior sites, paragraph 1
The outcome depends on which array is anterior, and neither the file nor the archive's metadata says which unit-id prefix is PMd (every unit's electrodes entry points to the PMd group for both prefixes). Under O1 range 44 of 72 prefix-1 units (0.611) and 49 of 70 prefix-2 units (0.700) pass the > 5 spikes/s rule (Fisher p = 0.293). With prefix 1 as PMd the anterior fraction is lower, not reproduced in all 12 readings; with prefix 2 as PMd it is higher without significance, partially reproduced in all 12 (Fisher p 0.293 to 1.000). Under O1 SD 4 of 72 and 4 of 70. Units whose range exceeds the 95th percentile of a condition-shuffled null are 7 of 72 and 6 of 70, and with condition labels shuffled 82 to 96 units pass the O1 range rule, so under O1 range these counts mostly measure noise. Median preparatory range is 6.58 spikes/s on the prefix-1 array and 10.19 on the prefix-2 array, and the mean preparatory rate 2.40 and 3.73 spikes/s.
How the test differed from the paper’s method
- AP location is the array, PMd anterior and M1 posterior (Supplemental Figure 3 legend); both mappings of the unit-id prefixes to the arrays are run.

Not tested in a notebook 4

4 data missing from the dataset

Blocked4

BlockedResults, Correlations between preparatory and peri-movement tuning, paragraph 1; Figure 4A to 4C· Data missing from the dataset
For the monkey A, B and J single-electrode datasets, mean preparatory to peri-movement tuning correlations are small (Pearson r = 0.21, 0.19 and 0.06, p = 0.01, 0.01 and 0.12; Spearman r = 0.20, 0.20 and 0.06).
Across the four datasets the mean correlation was positive (p=0.01, p=0.01, p=0.12, p=0.03) but only very modestly so (mean r = 0.21, 0.19, 0.06 and 0.13)
Results, Correlations between preparatory and peri-movement tuning, paragraph 1; Figure 4A to 4C
What the paper reports
- monkey A, mean r against zero: not named, p = 0.01
- monkey B, mean r against zero: not named, p = 0.01
- monkey J, mean r against zero: not named, p = 0.12
The dandiset holds one subject (Jenkins) recorded with two Utah arrays. The single-electrode datasets of monkeys A, B and J are not in it, and no other dandiset of these recordings was named by the paper or the dandiset.
BlockedResults, PD performance in a variety of spaces, paragraph 3; Figure 6A to 6C· Data missing from the dataset
For the monkey A, B and J single-electrode datasets as well, PD performance is best in the peri-movement space, including against the PCA-based EMG space.
performance was nearly always best for the space derived from the population-level peri-movement response. For every dataset, the highest level of performance was achieved
Results, PD performance in a variety of spaces, paragraph 3; Figure 6A to 6C
Needs the single-electrode datasets of monkeys A, B and J and their EMG, which are not in the dandiset.
BlockedResults, PD performance in a variety of spaces, paragraph 4; Figure 6E and legend· Data missing from the dataset
Across monkeys A, B and J, the peri-movement space gives the highest performance averaged across neurons, with the advantage growing as higher dimensionalities are allowed (limits 3, 6, 12); the task spaces are largely outperformed by the PCA-based kinematic and EMG spaces, which are in turn outperformed by the peri-movement space. Spaces marked by asterisks are significantly worse than the best space (p < 0.001).
Performance (averaged across neurons) was highest when the PD employed the space based on peri-movement neural activity ... the task spaces were largely outcompeted by the PCA-based kinematic and EMG spaces, which were in turn outperformed by the peri-movement space
Results, PD performance in a variety of spaces, paragraph 4; Figure 6E and legend
What the paper reports
- other spaces against the best space (asterisks): not named, p < 0.001 (n = neurons of monkeys A, B and J)
Figure 6E combines monkeys A, B and J with EMG spaces, none of which are in the dandiset. The variant on J-array without EMG is shown in the notebook (main reading); the peri-movement space has the highest mean at limits 3, 6 and 12 (-0.044), and the kinematic and initial-velocity spaces are not significantly worse (paired t p = 0.62 and 0.81).
BlockedResults, Comparison of anterior versus posterior sites, paragraph 1· Data missing from the dataset
For monkeys A, B and J, tuning correlation is not related to anterior-posterior location (r = -0.05, -0.03, -0.07; p = 0.82, 0.85, 0.52).
Regressing tuning correlation (as in Figure 4 ) against AP location yielded little or no effect ... r =-0.05, p =0.82; r =-0.03, p =0.85; r =-0.07, p =0.52
Results, Comparison of anterior versus posterior sites, paragraph 1
What the paper reports
- monkey A: linear regression, r = -0.05, p = 0.82
- monkey B: linear regression, r = -0.03, p = 0.85
- monkey J: linear regression, r = -0.07, p = 0.52
The single-electrode penetrations of monkeys A, B and J, and their AP coordinates, are not in the dandiset.

Findings about the data

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

Metadata errorEvery unit's `electrodes` entry in the train and test files points to an electrode of the PMd group (indices 0 to 94), though the unit ids carry two array prefixes (72 units with 1, 70 with 2) and the dandiset describes recordings from M1 and PMd. The electrode index equals the channel number in the id minus one for both prefixes, so the units of one array point to the wrong group and no unit points to an M1 electrode. Which prefix is PMd is not stated.
Description missingThe electrodes table has no positions or impedances (x, y, z and imp are NaN for all 192 electrodes), so no unit has an anterior-posterior location, which the paper's comparison of anterior and posterior sites uses.
Data missingThe dandiset holds 100 trials of 27 of the 108 maze conditions from the arrays of monkey J (1 to 4 trials per condition after the paper's delay rule), where the paper's J-array dataset has about 2166 trials per neuron. The single-electrode datasets of monkeys A, B and J and all EMG are not in it. Four claims cannot be tested, and the claims that were tested rest on condition means whose split-half reliability has a median of 0.20. With condition labels shuffled across trials, 82 to 96 of the 142 units pass the paper's > 5 spikes/s inclusion rule (read as the range), against 93 with the true labels; the paper justified that threshold by its high trial count.
Selection not documentedThe dandiset says the data are "limited to 100 train trials and 100 test trials" but not how the 27 conditions and 100 trials were chosen, whether this session is the paper's J-array dataset, or whether the trials were filtered by the paper's velocity-profile rule. 22 of the 100 train trials have delays of 400 ms or less, which the paper excludes.
OtherThe test file's 100 segments (0.7 s, movement onset at 0.25 s) have no condition, go cue, target or behaviour, and hold only the 107 held-in units, so they cannot enter condition-wise analyses. None duplicates a train trial (smallest summed difference in spike counts 117).
Methods not availableQuestion for the paper's authors. The inclusion rule ">5 spikes/s of preparatory tuning" (10 spikes/s for single electrodes) does not define tuning. Read as the range across conditions it keeps 93 of 142 units here, as the SD 8; outcomes of fig4-broad-near-zero, fig4e, fig6-maze-endpoint-speed-space-poor and fig6-dimensionality-trend differ between the two. Was it the range, the SD, or another measure, and was the same measure used for the 1.5 times peri-movement rule?
Methods not availableQuestion for the paper's authors. How was the maze-task "proto-trial" chosen (a trial best matching the condition mean, or the mean itself), over what window were velocity profiles correlated, and how was movement onset detected (Churchland et al. 2006c is cited)? With the condition mean as proto-trial or a speed-threshold onset the fig6d result is partially reproduced here, with a proto-trial and the dataset's onset reproduced.
Methods not availableQuestion for the paper's authors. For the task spaces, how was the end of the reach set for the endpoints, were the three segments of equal length or duration, was "peak horizontal and vertical velocity" signed, and were task-space axes scaled before the PD consistency analysis of Figure 7D? Did the upper-limit space of the supplement include the fitted neuron, and did the peri-movement T matrix use all recorded neurons or only those passing the inclusion rule? Under the SD reading of tuning the fig6d result here depends on it.

Notebooks

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

Reproduction Preparatory and peri-movement tuning in the maze-task array data (Churchland et al. 2010)

Tests the findings of Churchland, Cunningham, Kaufman, Ryu and Shenoy (2010) on dandiset 000140 (MC_Maze_Small), which holds 100 trials of 27 maze conditions from the two Utah arrays of monkey J. It correlates preparatory and peri-movement tuning, compares how well a preferred direction in the peri-movement, kinematic and task spaces captures preparatory tuning, runs the paper's controls, compares the two arrays, and repeats the paper's simulation of a rotational linear system. The single-electrode datasets and EMG of the paper are not in the dandiset. All files

Reproducing Churchland et al. (2010) from dandiset 000140
reproduce_churchland_2010.ipynbExecuted, with outputs · Source

Verification

Round 1
2026-10-29
Agreed with 16 of 20 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook), verifier-1. Reviewed: The handover of the first run, the executed notebook with the conclusions and summary removed, the helper module and the claims file, with the paper under paper/ and the archive's API.
- fig4-broad-near-zero: the verifier’s outcome was “Not reproduced”. Every range (O1a) reading fails the IQR >= 0.3 test (IQR 0.17-0.29). The only passes are the SD readings, with n = 3-5 units. An IQR near 0.26 is what pure noise gives with 27 conditions.
- fig6d-perimovement-best-j-array: the verifier’s outcome was “Partially reproduced”. 64 runs reproduced and 32 partial (initial vel. >= peri under O3/O4 a,b / b,a / b,b). The criterion needs every reading to hold. Under O1a every space has negative performance, and the gaps are about 0.005; peri vs kinematic p=0.62 and peri vs initial vel. p=0.81. No per-reading SEs or paired tests were output.
- fig7c-anterior-posterior-performance: the verifier’s outcome was “Blocked”. The upper limit is <= 0 in 8/10 rows, so the normalised measure the criterion uses does not exist. Raw 6-D performance is about -1.1 to -1.5 at both sites. Which array is anterior is also unverified, and the swap promised in the notebook is missing from this cell (problem 4).
- prep-activity-more-anterior: the verifier’s outcome was “Blocked”. With prefix 1 taken as PMd, PMd 0.61 vs M1 0.70 (p=0.29) is not_reproduced; with the arrays swapped it would be partially_reproduced. Neither the file nor the API metadata says which array is which, and every unit's electrode index points to the PMd group for both prefixes, so the prefix 1 = PMd reading has no support in the file (problem 4).
Round 2
2026-10-29Agreed with 20 of 20 outcomes. Independent agent run by CatalystNeuro (same harness, separate context, did not write the notebook), verifier-2. Reviewed: The full executed notebook of the first revision, with its conclusions and summary, the helper module, the README, the attempt record and the claims file, with the paper under paper/.

Method differences

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

Assumed O1 is the range across conditions (see fig4-broad-near-zero).fig4e-correlation-time-course fig6-dimensionality-trend
27 conditions with 1 to 4 trials each after the delay rule, against 108 conditions and about 20 trials each in the paper's J-array dataset.fig4d-mean-correlation-j-array
Spike trains smoothed with a Gaussian of SD 20 ms before averaging (Supplemental Experimental Procedures, Computing the mean firing rate versus time; the width is read as the SD, a standard choice).fig4d-mean-correlation-j-array
Assumed O1 is the range of the preparatory rate across conditions, as the Methods describe the inclusion rule as a rate difference ("Firing rate modulation was typically lower ... >5 spikes/s of preparatory tuning"); under the SD reading the claim is reproduced, on 3 to 5 units.fig4-broad-near-zero
The time axis uses go-locked means from -200 to +200 ms and movement-locked means from -400 to +400 ms, in 10 ms steps (Supplemental Experimental Procedures, the spliced trace); the judgement uses the movement-locked part.fig4e-correlation-time-course
Each unit is normalised by its maximum condition-mean rate over the go-locked and movement-locked spans stored (-300 to +200 ms and -400 to +900 ms); the legend says "normalizing each neuron's activity by its maximum rate".suppfig1-population-averages-lose-tuning
Assumed O1 range (see fig4-broad-near-zero). Under O1 SD with the T matrix of the included units the claim is not reproduced; under O1 range every reading gives reproduced or partial.fig6d-perimovement-best-j-array
The units of the peri-movement T matrix are read two ways, all recorded units (main) and the units passing the inclusion rule. The Methods say each row "contained data from all neurons except the neuron whose preparatory activity we were trying to capture" and that "Analysis of preparatory responses was restricted to neurons with at least 10 spikes/s of preparatory tuning".fig6d-perimovement-best-j-array
The upper limit uses the preparatory matrix of all units except the fitted one, as the peri-movement space does (Methods, PCA-based spaces); the supplement says "all neurons".fig6d-perimovement-best-j-array
Task spaces follow the Methods' list (Task spaces). Details they leave open, set here and stated in maze_prep.py, are the end of the reach (last time within 800 ms of onset with speed at least 10% of peak), the peak velocity components (the signed value of largest magnitude), and the segments (equal path length, with equal duration as a variant, which gives the same outcome).fig6d-perimovement-best-j-array
Acceleration for the kinematic space is the derivative of the dataset's hand velocity after a 20 ms Gaussian, a standard choice.fig6d-perimovement-best-j-array
Assumed O1 is the range across conditions (see fig4-broad-near-zero); under the SD reading with the T matrix of all recorded units the claim is reproduced.fig6-maze-endpoint-speed-space-poor
Assumed O1 SD (O1b). Under the O1 range reading adopted elsewhere (see fig4-broad-near-zero) performance at -150 ms is below zero and the ratio is undefined, so the outcome rests on the O1b readings (7 or 8 units) alone.fig7b-window-start-slide
Start times -150 to +100 ms in 50 ms steps, end fixed at +800 ms, best k up to 14 for each start (Figure 7B legend, "the best dimensionality for that dataset/start-time").fig7b-window-start-slide
The electrodes have no positions, so AP location is the array, PMd anterior and M1 posterior (Supplemental Figure 3 legend). Regression on a 0/1 location, the two-group comparison the criterion names as the variant for array identity.ap-location-no-effect-j-array
Assumed the array of a unit is the first digit of its id, 1 for PMd and 2 for M1, as units with prefix 1 map onto the PMd electrodes of the table; every unit's electrodes entry points to the PMd group. Swapping the arrays changes only the sign of r.ap-location-no-effect-j-array
AP halves are the two arrays (see ap-location-no-effect-j-array), not a median split of positions.fig7c-anterior-posterior-performance
O1 range, O3a and O4a (see fig6d-perimovement-best-j-array); 20 random halves of the size of the prefix-1 array.fig7c-anterior-posterior-performance
AP location is the array, PMd anterior and M1 posterior (Supplemental Figure 3 legend); both mappings of the unit-id prefixes to the arrays are run.prep-activity-more-anterior
The paper does not give the scale of task-space axes, which sets a PD's direction when units are mixed (mm and mm/s); raw and z-scored axes are both run and agree on the segment-directions comparison in 11 of 12 pairs (they differ under O1 SD with O3a and O4b).fig7d-pd-consistency
Draws in which the unit's rate is the same in all drawn conditions (a zero PD) are dropped; 100 repeats per unit, the claim's standard choice.fig7d-pd-consistency

Provenance

Attempt record: 000140-churchland-2010-01
Paper: Cortical preparatory activity: representation of movement or first cog in a dynamical machine?
Date: 2026-10-29
Made by: An agent, run by CatalystNeuro
Dataset: DANDI:000140
Version: 0.220113.0408 (doi:10.48324/dandi.000140/0.220113.0408)
Run on: GitHub Actions
Harness: Claude Code
Model: claude-opus-5-5
Protocol: pilot-12

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.

Other datasets that name the same paper

← DANDI:000139 DANDI:000147 →