Claims 22
From the paper Cortical preparatory activity: representation of movement or first cog in a dynamical machine? · claims file · checked 2026-10-22: ready
- recordings-jarray
The J-array dataset holds 146 single- and multi-unit isolations from two 96-electrode arrays, recorded on the maze task with 108 conditions, with a mean of 2166 trials per neuron.
(27 and 108 conditions for ‘monkey J’ and ‘monkey J-array’ datasets) ... a pair of implanted 96-electrode arrays (J-array dataset, 146 single- and multi-unit isolations) ... High trial counts (mean trials/neuron = 311, 492, 388 and 2166 for monkey A, B, J, J-array)
Results, Behavior and recordings, paragraph 1; Methods, Neural recordings and datasetsHow it will be judged
Counts from the dataset's units, conditions and trials. The trial count can be read as (a) all successful trials or (b) those kept by M1 (delay > 400 ms and the proto-trial selection); the Methods print 2155 where the Results print 2166, both within the tolerance. The condition count has tolerance 0 because it is a design count, not a measurement. Reproduced when all three are within tolerance under one reading.
- recordings-single-electrode
The single-electrode datasets hold 64, 74 and 172 single units for monkeys A, B and J, with means of 311, 492 and 388 trials per neuron.
High trial counts (mean trials/neuron = 311, 492, 388 and 2166 for monkey A, B, J, J-array) ... A total of 310 single-electrode isolations (64, 74, and 172 single-unit isolations for monkeys A, B and J) were recorded from caudal PMd, surface M1 and sulcal M1.
Results, Behavior and recordings, paragraph 1; Supplemental Experimental Procedures, Neural and EMG recordingsHow it will be judged
Testable only if the dataset holds the speed task or the single-electrode maze recordings. Reproduced when every count the dataset can give is within tolerance.
- fig2b-example-concordant-tuning
In the speed task one example neuron keeps its preparatory preference (reaches up and left) into the peri-movement epoch, with a preparatory to peri-movement tuning correlation of 0.72.
This preference is shared between preparatory and peri-movement epochs. This agreement is not perfect (two green traces surpass the black trace) but the overall correlation is quite high ( r =0.72, see section on correlations below for details).
Results, Neural responses during preparation and movement, paragraph 2; Figure 2BHow it will be judged
Speed task only (monkey A or B); the paper does not name the neuron. Per neuron, the tuning correlation of claim fig4abc-mean-correlation-single-electrode. The paper gives no measure of "prefers reaches ending up and leftwards"; the Figure 2B legend shows the near distance and fast speed and shades traces by the rate 50 ms before the go cue. It is read as follows. Among those 7 conditions, the one with the highest rate 50 ms before the go cue has its target in the upper-left quadrant. Reproduced when at least one speed-task neuron has r within tolerance of 0.72 and that preference; blocked if the dataset holds no speed-task data.
- fig2ce-example-opposing-tuning
Three other speed-task neurons with clear preparatory tuning do not keep their preferences into the peri-movement epoch; their preparatory and peri-movement tuning are negatively correlated, even for straight reaches at one distance and speed.
Despite clear preparatory tuning, preferences are not maintained between the preparatory and peri-movement epochs (color ordering differs between epochs). ... These neurons show negative correlations between preparatory and peri-movement tuning.
Results, Neural responses during preparation and movement, paragraph 3; Figure 2C-E and legendHow it will be judged
Speed task only, neurons passing M6, tuning correlation as in M4. The paper does not say over which conditions the negative correlations of these neurons were computed. Readings (F1) the 7 far, fast conditions shown in the Figure 2C-E legend; (F2) all 28 conditions, as in Figure 4. Reproduced when at least three neurons have negative correlations under one reading; blocked if the dataset holds no speed-task data.
- fig3-maze-most-not-maintained
In the maze task some neurons keep similar tuning between the preparatory and peri-movement epochs, but most do not.
Some neurons maintained similar tuning between preparatory and peri-movement epochs (panel A ) but most did not.
Results, Neural responses during preparation and movement, paragraph 4; Figure 3How it will be judged
J-array (and J if present), neurons selected as for Figure 4 (M6, both readings), tuning correlation as in M4. The paper sets no threshold for "similar tuning"; the one value it calls "quite high" is r = 0.72 (Figure 2B). Reproduced when at least one neuron has r of 0.72 or more and more than half of the neurons have r below 0.72.
- fig4d-mean-correlation-jarray
In the J-array dataset the correlation between a neuron's preparatory and peri-movement tuning is positive on average but small (mean 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 4DHow it will be judged
Per neuron passing M6 (preparatory tuning > 5 spikes/s and peri-movement tuning at least 1.5 times the preparatory tuning, readings S1 and S2), the Pearson and Spearman correlation over the 108 conditions between the preparatory rates and the peri-movement rates of M4, trials by M1 (T1, T2). The paper does not name the test of the mean against zero. Readings (P1) one-sample two-sided t-test of the r values; (P2) the same on Fisher z-transformed r; (P3) two-sided Wilcoxon signed-rank test. Reproduced when, under one combination of readings, both means are within tolerance and the test gives p < 0.05 with the mean positive; partially reproduced when the mean is positive with p < 0.05 under one combination but no combination puts both means within tolerance.
- fig4abc-mean-correlation-single-electrode
In the single-electrode datasets (monkeys A, B and J) the mean tuning correlation is 0.21, 0.19 and 0.06 (p = 0.01, 0.01 and 0.12), with mean Spearman correlations of 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). 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 4A-CHow it will be judged
As fig4d-mean-correlation-jarray, with the 10 spikes/s preparatory threshold, the speed task peri-movement window as in M4, and for monkey J the four 27-condition sets analysed as separate datasets (Methods, Task design). Reproduced per dataset the archive holds; blocked for datasets it does not hold.
- fig4-no-separate-populations
The distributions of tuning correlations give little evidence (except perhaps weakly for monkey B) of one population with concordant tuning (r near 1) and another with opposing tuning (r near -1); most neurons have weakly correlated tuning, with a broad range of correlations centred near zero.
Even more disconcertingly, different neurons show markedly different relationships between preparatory and peri-movement activity, with a broad range of correlations centered near zero. ... Yet there is little evidence in Figure 4 (except perhaps weakly in panel B ) for one population of concordant tuning ( r ≈ 1) and another of opposing tuning ( r ≈ -1). Instead, most neurons have weakly correlated tuning.
Results, Correlations between preparatory and peri-movement tuning, paragraphs 3 and 4; Figure 4A-DHow it will be judged
Per-neuron correlations from fig4d (and fig4abc for datasets the archive holds). A claim of absence judged by eye in the paper, which names no test and no threshold. Readings (U1) Hartigan's dip test of unimodality gives p >= 0.05; (U2) more neurons have |r| < 0.5 than have |r| >= 0.5. Reproduced when both hold, not reproduced when neither holds, and methods_underspecified when they disagree; the dip statistic and the share of |r| >= 0.8 are reported either way.
- fig4e-correlation-time-course
The correlation between preparatory tuning and peri-movement tuning measured at single time points begins to decline about 200 ms before movement onset and is near zero by movement onset.
The correlation with preparatory activity began to decline ∼200 ms prior to movement onset, and reached values near zero by movement onset ( Figure 4E ).
Results, Correlations between preparatory and peri-movement tuning, paragraph 2; Figure 4E and legendHow it will be judged
Neurons as in fig4d. Per neuron, the M4 preparatory rates correlated over conditions with the rate at single times, smoothed with a 20 ms Gaussian (Figure 4E legend), every 10 ms from 400 ms before to 200 ms after movement onset; the mean over neurons at each time. The paper used rates spliced from go-cue and movement-onset locked averages, with the gap placed to minimise discontinuities (Supplemental Experimental Procedures); the placement is open. Readings (A1) movement-onset locked rates throughout; (A2) go-locked rates up to 100 ms after the go cue and movement-locked from 300 ms before onset, interpolated linearly in between where they overlap. The paper gives no test; "near zero" and "began to decline" are read as (i) the mean correlation at movement onset is not significantly above zero by the tests of fig4d (P1 to P3), and (ii) the mean correlation at movement onset is below that at 200 ms before onset, and more than half the drop from 400 ms before onset to onset happens after 200 ms before onset. Reproduced when (i) and (ii) hold.
- suppfig1-population-averages
In population averages sorted by each neuron's preparatory preference, the weak agreement between preparatory and peri-movement tuning dominates early, but the separation shrinks with time and shows little relation 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 ( Supp. Fig. 1 ).
Results, Correlations between preparatory and peri-movement tuning, paragraph 3; Supplemental Figure 1 and legendHow it will be judged
Neurons as in fig4d (Supplemental Figure 1 legend). Per neuron, the conditions with the strongest and weakest preparatory rate and two intermediate ones (open; read as the conditions at one third and two thirds of the rank order), each neuron's rates divided by its maximum rate, then averaged over neurons. Separation is the strongest-minus-weakest average. The paper gives no threshold for "dramatically smaller"; readings (R1) separation at movement onset less than half its mean over the preparatory epoch; (R2) less than a quarter. Reproduced when the separation in the preparatory epoch is positive and the condition holds under one reading.
- fig6-peri-movement-space-best
For every dataset the highest cross-validated PD performance in accounting for preparatory tuning is reached when the PD is expressed in the space built by PCA from the population peri-movement response, compared with task spaces and PCA-based kinematic and EMG 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, paragraph 3; Figure 6A-DHow it will be judged
J-array (and other datasets the archive holds). Neurons passing the preparatory part of M6 (S1, S2), regardless of peri-movement tuning (Methods, Correlations, last sentence). PD performance by M7 in (a) the peri-movement space (M5) for each k (M8, K1 and K2); (b) the kinematic space, PCA of a T-matrix of horizontal and vertical position, velocity and acceleration from 150 ms before movement onset, each variable scaled to unit range over times and conditions (Methods, PCA-based spaces), for each k; (c) every task space of M9 (V1, V2, G1, G2). EMG is not recorded for J-array. The end of the kinematic window is open ("until after the reach"); readings (W1) 800 ms after onset, as for the neural window; (W2) 600 ms, the longest reach duration in Figure 1 legend. The Methods say "each neuron contributed one kinematic variable"; readings (D1) the six variables once each; (D2) the six variables repeated in turn to make as many column blocks as there are neurons. Reproduced when, under every combination, the best mean performance over k of the peri-movement space exceeds the best of every other space; the outcome is methods_underspecified if the combinations disagree.
- fig6-each-dimensionality
Considering each dimensionality separately, PD performance is almost always higher in the peri-movement space than in other spaces of the same dimensionality; for the maze task the 3-D endpoints and peak speed space does not perform well.
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. 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 6A-DHow it will be judged
Performances from fig6-peri-movement-space-best. Comparisons are each task space against the peri-movement space at the task space's dimensionality, and the kinematic space against the peri-movement space at each k. "Almost always" has no threshold in the paper. Readings (L1) all comparisons but at most one, the paper naming one class of exception (3-D spaces on the speed task); (L2) at least 90% of comparisons. "Did not perform well" for the 3-D endpoints + maximum speed space is read as below the 3-D peri-movement space. Reproduced on J-array when that space falls below the 3-D peri-movement space and the peri-movement space wins under L1 and L2; the outcome is methods_underspecified if L1 and L2 disagree.
- fig6-performance-vs-dimensionality
PD performance in PCA-based spaces generally improves with dimensionality, peaks in a middle range, and declines beyond about 10 dimensions through over-fitting.
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-DHow it will be judged
Peri-movement space mean performance over k = 1 to 14 (K1). The paper gives no test; "improved with dimensionality" and "declined at dimensionalities >10" are read as performance at the best k higher than at k = 1, the best k at 10 or below, and performance at k = 14 below that at the best k. Reproduced when all three hold.
- prep-dimensionality
Preparatory activity is moderately high-dimensional; depending on the dataset, 7 to 10 dimensions capture 90% of its variance across conditions.
This T-matrix also allows us to estimate the dimensionality of preparatory activity ... 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 2How it will be judged
The c x n matrix of M4 preparatory rates (one row per condition, one column per neuron, not normalised, columns mean-centred), PCA, the smallest number of components whose cumulative variance reaches 90%. The range 7 to 10 widened by 10% of each bound gives 6.3 to 11, so 7 to 11 components. Which neurons enter is open; readings (N1) neurons passing the preparatory part of M6 (S1, S2); (N2) all neurons. Reproduced when the count is within the range under one reading.
- fig6e-summary-across-datasets
Averaged over monkeys A, B and J (J-array excluded, no EMG), PD performance is highest in the peri-movement space at every dimensionality limit, the advantage grows with dimensionality, and task spaces are largely outperformed by the kinematic and EMG spaces, which are in turn outperformed by the peri-movement space.
Performance (averaged across neurons) was highest when the PD employed the space based on peri-movement neural activity. This performance differential increased at higher dimensionalities, indicating that the peri-movement space better captured higher-dimensional aspects of the data.
Results, PD performance in a variety of spaces, paragraph 5; Figure 6E and legendHow it will be judged
Needs the speed task, the single-electrode J data and EMG; J-array is excluded by the paper. Limits of 3, 6 and 12 dimensions; for PCA-based spaces the best k up to the limit; only spaces defined for all three datasets (endpoints + maximum speed included, targets + instructed speed not); performance averaged over neurons, SE over neurons. The test behind the asterisks is not named; readings (Q1) paired t-test over neurons; (Q2) Wilcoxon signed-rank over neurons. The paper gives no threshold for "increased" or "largely"; they are read as the gap between the peri-movement space and the next best space being larger at the 12-dimension limit than at the 3-dimension limit, and the kinematic and EMG spaces each exceeding more than half of the task spaces at the 12-dimension limit. Reproduced when the peri-movement space is best under all three limits, no other space is significantly better, and both readings hold; blocked if the archive lacks those datasets.
- fig7a-shuffled-space-poor
A control space built from peri-movement responses in which each neuron's peri-movement pattern was inverted on a random half of the conditions performs poorly, so the trailing end of preparatory activity does not explain the peri-movement space's success.
The PCA-based ‘shuffled-space’ performed poorly ( Figure 7A ); any contribution from the trailing end of preparatory activity (preserved despite shuffling) is insufficient to allow good performance.
Results, Controls, paragraph 1; Figure 7A and legendHow it will be judged
As fig6-peri-movement-space-best, with the T-matrix shuffled per neuron and condition with probability 0.5 (Figure 7A legend), activity kept up to 150 ms before movement onset and later activity reflected about the rate at that time (r'(t) = 2 r(-150 ms) - r(t)). The number of shuffles is not given; 10 shuffles with the mean performance reported and the spread shown, a standard choice. "Performed poorly" has no threshold; readings (H1) below the unshuffled peri-movement space at every k; (H2) H1 and its best over k below the best of the kinematic space. Reproduced when both hold; methods_underspecified when only H1 holds.
- fig7b-window-start-slide
Moving the start of the peri-movement window from 150 ms before movement onset to movement onset costs little PD performance; at movement onset performance is 97% (monkey J) and 94% (monkeys A, B) of its original value.
Finally, the start of the peri-movement window was 150 ms before movement onset but could be slid later in time with little decline in performance ( Figure 7B ). 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 and legendHow it will be judged
As fig6-peri-movement-space-best, the peri-movement window starting at -150 ms and at 0 ms from movement onset, its end fixed; performance at the best k for each start (Figure 7B legend). "Monkey J" may mean (J1) the single-electrode J dataset only, which the archive may not hold, or (J2) both J datasets; under J2 the J-array ratio is compared with 0.97. Reproduced when a ratio the dataset can give is within tolerance.
- ap-location-no-effect-on-correlation
The weak correlation between preparatory and peri-movement tuning does not depend on anterior-posterior recording location; regressing tuning correlation on AP location gives r = 0.20, p = 0.27 for J-array and no significant effect in any dataset.
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; and r =0.20, p =0.27 for the four datasets.
Results, Comparison of anterior versus posterior sites, paragraph 1How it will be judged
A claim of absence, judged on the paper's test, a linear regression (Pearson r and its p) of each neuron's tuning correlation (fig4d) on its AP location. Needs a per-unit AP location; for J-array, the electrode's position along the anterior-posterior axis of the arrays if the dataset gives it, else the array (PMd anterior, M1 posterior) as a two-level variant. Reproduced when p >= 0.05; not reproduced when p < 0.05; r reported either way.
- fig7c-anterior-posterior-performance
The peri-movement space captures preparatory tuning at both anterior and posterior sites, with 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 legendHow it will be judged
As Figure 7C legend, a six-dimensional peri-movement space; neurons split at the median AP location (anterior at or above, posterior at or below); performance normalised by its upper limit, the performance of a PCA space from the preparatory activity of all neurons (Supplemental Experimental Procedures), and compared with random subdivisions of the same sizes. Whether the space for each half is built from that half only is open; readings (B1) from that half's neurons; (B2) from all neurons. Reproduced when normalised performance for both halves is above zero and within the spread of the random subdivisions or above it, and anterior is higher than posterior.
- fig7d-pd-consistency
When the PD is measured on two random sets of 25 conditions, it is most stable in the peri-movement space, for every comparison space and all four datasets.
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, paragraph 2; Figure 7D and legendHow it will be judged
Per neuron, 25 conditions drawn with replacement, PD fit by M7 regression (unit length), another 25 drawn, PD refit; consistency is the dot product of the two PDs, averaged over neurons (Results, Stability of the PD). The number of draws per neuron is not given; 100 draws averaged, a standard choice. Comparison spaces for the maze task (Figure 7D legend) endpoints + maximum speed (3-D), segment directions (6-D), the kinematic space at its best k from Figure 6 for this dataset; the peri-movement space at the same dimensionality as each. Reproduced when the peri-movement space has the higher mean consistency in every comparison.
- fig8-model-weak-correlation
In a simulated linear dynamical system (a harmonic oscillator held at 25 random initial states, then released), the preparatory to peri-movement tuning correlation of a unit is low (0.08), and a given preparatory rate can lead to opposing peri-movement patterns.
The correlation of preparatory activity with peri-movement activity (averaged over the gray interval) was 0.08.
Results, Tuning from a dynamical systems perspective, paragraph 2; Figure 8B-C and legendHow it will be judged
A simulation, not a test on recorded data. Two units, x(t+1) = W x(t) with W as printed in the Figure 8 legend, [[1, -2 pi/360], [2 pi/360, 1]] (the exact rotation shown as a variant), 25 random initial states. Open details, the distribution of initial states and the gray interval; readings (I1) uniform in [-1, 1] squared; (I2) standard normal; (J1) the first 360 steps after release; (J2) the first 180 steps. The seed is unknown, so the value is judged against the spread over 1000 draws of 25 states. Reproduced when 0.08 lies within the central 95% of that spread under one combination.
- fig8-model-pd-perfect
In the same simulated system, preparatory tuning is captured perfectly by a PD in the two-dimensional space found by PCA of the peri-movement activity.
The PD points towards the pattern recruited by the activity of unit one (panel A ), and perfectly captures preparatory ‘tuning’.
Results, Tuning from a dynamical systems perspective, paragraph 3; Figure 8D and legend; Supplemental TextHow it will be judged
A simulation, not a test on recorded data. As fig8-model-weak-correlation; PCA of the 25 x 2t matrix of peri-movement activity, k = 2, PD fit of unit 1's initial rates by M7 without cross-validation and with it. Reproduced when the fit is exact to numerical precision (R squared and leave-one-out performance above 0.999) under every reading.
Pipeline
- Paper
- done (2 of 2)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none