DANDI:001605 · Encore Science

Claims 40

From the paper Neural substrate of dynamic Bayesian inference in the cerebral cortex · claims file · checked 2026-10-21: ready

session-counts
Imaging data come from 94 sessions of 8 mice (2, 10, 5, 14, 15, 15, 12 and 21 sessions for mice 1 to 8); PPC layers 2, 3 and 5 were imaged in 17, 15 and 14 sessions and PM layers 2, 3 and 5 in 16, 15 and 17.
In total, we analyzed data from 94 sessions in 8 mice (mouse 1, 2 sessions; mouse 2, 10 sessions; mouse 3, 5 sessions; mouse 4, 14 sessions; mouse 5, 15 sessions; mouse 6, 15 sessions; mouse 7, 12 sessions; mouse 8, 21 sessions)
Online Methods, Behavioral analysis; Results, Goal-distance representation in PPC and PM, paragraph 1
How it will be judged
Counts of named sessions and animals, so the tolerance is 0 for each, and the sessions per mouse must match the list as a multiset (the dataset's subject labels need not follow the paper's numbering). Reproduced when all counts match; partially reproduced when the totals match and at most two per-group counts differ by one.
neuron-counts
41,031 neurons were imaged, 8,371, 7,108 and 4,859 in PPC layers 2, 3 and 5 and 7,788, 7,005 and 5,900 in PM; these are active cells, 99.6% of the manually selected ROIs.
In total, we imaged the activity of 41,031 neurons in 94 sessions (PPC ... 8,371, 7,108 and 4,859 neurons from 17, 15 and 14 sessions in layers 2, 3 and 5, respectively; PM ... 7,788, 7,005 and 5,900 neurons from 16, 15 and 17 sessions, respectively)
Results, Goal-distance representation in PPC and PM, paragraph 1; Supplementary Table 1
How it will be judged
The per-session averages of Supplementary Table 1 (active cells 492, 474, 347, 487, 467, 347) times the session counts give these totals, so they count active cells. Readings of what the dataset's ROIs are, (a) all ROIs, (b) ROIs with a dR/R excursion beyond 3 s.d. of the session trace (the paper's "signal beyond 3σ"). Reproduced when, under one reading, all seven counts are within tolerance.
behaviour-timing
Mice got the reward 7.4 ± 0.1 s after the trial start and spent 860 ± 10 ms in each no-sound zone of the intermittent conditions (mean ± s.e.m. over 94 sessions).
Mice got the reward in 7.4 ± 0.1 s (mean ± s.e.m., n = 94). In intermittent conditions, mice spent 860 ± 10 ms (n = 94) in each no-sound zone.
Online Methods, Behavioral training, paragraph 5
How it will be judged
Tolerances are the stated s.e.m. Per session, the mean over the analysed trials (first 150, without trials over 20 s or error trials), then the mean over sessions. Which no-sound zones count is not given; readings (a) every complete no-sound zone after the first 33 cm and before the goal, (b) every no-sound zone the mouse enters, including the one around the goal in intermittent 1 up to the reward. Reproduced when the reward time and, under one reading, the zone time are within tolerance.
lick-goal-correlation
In every session the start-to-first-lick distance correlated with the start-to-goal distance, with r from 0.258 to 0.822 over the 94 sessions (r = 0.798 in the example session).
Mice increased licking frequency as they approached the goal (r = 0.798, P = 2.11 × 10−34; Fig. 1b; 94 sessions of 8 mice ... r = 0.258 to 0.822, P = 0.00229 to 9.05 × 10−38)
Results, Licking indicates action-dependent distance estimation, paragraph 2; Figure 1b; Online Methods, Behavioral analysis
How it will be judged
The Results do not name the two variables of r. They are read as those of the Methods' session selection, since every reported P is below its 0.01 threshold. The Methods define that correlation as start-to-lick distance against start-to-goal distance, first lick meaning the first lick more than 33 cm from the start, over the analysed trials of all conditions. The coefficient is not named; readings (a) Pearson, (b) Spearman. Reproduced when, under one reading, every session has r > 0 with P < 0.01 and all r lie within 0.232 to 0.904 (the range widened by 10% of each bound).
fig1c-licking-without-sound
Licking increased as mice approached the goal in all conditions, and in both intermittent conditions it increased even within the no-sound zones (41.9-25.1 and 8.4-0 cm in intermittent 1, 25.1-8.4 cm in intermittent 2).
In intermittent conditions, anticipatory licking increased even when the sound was omitted (one-sided Wilcoxon signed-rank test, P = 4.87 × 10−8 to 1.22 × 10−15; Fig. 1c)
Figure 1c; Results, Licking indicates action-dependent distance estimation, paragraph 2
How it will be judged
Licks per trial in goal-distance bins, mean per session, for each condition (Fig. 1c). The text does not state the bins or the pairs compared; they are read from the figure. The boxes are 4.2 cm wide (32 from 134 to 0 cm), and each of the four brackets compares a bin with the bin 8.4 cm (two boxes) farther from the goal. The tested bins are 29.3-25.1, 20.9-16.7, 12.6-8.4 and 4.2-0 cm. The text confirms this. Its no-sound range, P = 4.87e-8 to 1.22e-15, is exactly the set of the first and fourth intermittent 1 values and the second and third intermittent 2 values, the comparisons whose tested bin is silent. Bin edges at multiples of 4.2 cm from the goal are a standard choice read from the figure. One-sided signed-rank test over the 94 sessions (tested bin higher), at the Bonferroni threshold P < 0.01 (Statistics; the legend marks P < 1e-7). Reproduced when licking rises with P < 0.01 in the four no-sound comparisons (intermittent 1, first and fourth; intermittent 2, second and third) and in all four continuous comparisons. The other comparisons are shown.
supp1-licking-speed
During no-sound zones licking rose more steeply in fast-locomotion trials than in slow ones, in both intermittent conditions.
Slope of licking ... was significantly steeper in fast trials than slow trials, suggesting that mice use action-dependent distance estimation
Supplementary Figure 1; Results, Licking indicates action-dependent distance estimation, paragraph 2
How it will be judged
Per trial, lick frequency in a no-sound zone regressed on time (Lick(t) = b0 + b1 t, Supplementary Fig. 1); trials split into equal slow and fast halves per session (O9); per session, mean slope of slow and of fast trials; signed-rank test over sessions per condition. Reproduced when, under every reading of O9, fast slopes exceed slow with P < 0.01 in both intermittent conditions.
fig1d-muscimol-correlation
Silencing PPC with muscimol lowered the correlation between start-to-lick and start-to-goal distance, making lick initiation more variable.
Correlation coefficients decreased after muscimol injection, indicating that silencing of PPC made lick initiation more variable (Mann-Whitney U test, P = 3.08 × 10−4, z = 3.61)
Figure 1d; Results, Licking indicates action-dependent distance estimation, paragraph 3
How it will be judged
Session correlations as in lick-goal-correlation, muscimol sessions against PBS sessions (no-injection sessions are shown in Fig. 1d but not tested), two-sided Mann-Whitney U. Reproduced when muscimol correlations are lower with P < 0.05.
fig1e-muscimol-variance
After muscimol the distance between the first lick and the goal was more variable than after PBS, in all three conditions.
Variance of the distance was compared between PBS and muscimol-injected trials
Figure 1e; Results, Licking indicates action-dependent distance estimation, paragraph 3
How it will be judged
First-lick-to-goal distance per trial, PBS against muscimol trials, Ansari-Bradley test per condition. Reproduced when muscimol trials have the larger spread with P < 1e-14 (the legend's threshold) in all three conditions.
fig1fg-muscimol-no-sound-licking
After muscimol, licking no longer increased significantly in the no-sound zones of the intermittent conditions, while after PBS it did. The increase of licks between adjacent bins in those no-sound zones was smaller after muscimol than after PBS.
The effect of PPC silencing was most prominent in intermittent conditions, with no increase in licking in no-sound zones (one-sided Wilcoxon signed-rank test, P = 0.485 to 0.0212; Fig. 1f; Mann-Whitney U-test, P = 0.00761 and 4.84 × 10−6; Fig. 1g)
Figure 1f,g; Results, Licking indicates action-dependent distance estimation, paragraph 3
How it will be judged
Fig. 1f, licks per trial in 4.2-cm bins and the same four comparisons as fig1c-licking-without-sound (each tested bin against the bin 8.4 cm farther from the goal; the figure draws the same brackets), separately for the 12 muscimol and the 12 PBS sessions, one-sided signed-rank test over sessions, P < 0.01 (legend). The text's muscimol range for no-sound zones, P = 0.485 to 0.0212, is again the first and fourth intermittent 1 values and the second and third intermittent 2 values. Fig. 1g, per trial, the change of lick count from the 8.4-cm bin 25.1-16.7 cm to 16.7-8.4 cm and from 16.7-8.4 cm to 8.4-0 cm (the legend's 25-17, 17-8.4 and 8.4-0 cm), PBS against muscimol trials, Mann-Whitney U, P < 0.01. Reproduced when none of the four no-sound comparisons has P < 0.01 after muscimol, all four have P < 0.01 after PBS (the paper's largest is 0.00464), and the Fig. 1g increase is larger after PBS with P < 0.01 for intermittent 1 at 17-8.4 to 8.4-0 cm and intermittent 2 at 25-17 to 17-8.4 cm.
supp2d-muscimol-overrun
Silencing PPC increased overrun, the start of licking more than 8.3 cm past the goal, most in the intermittent 1 condition.
Silencing of PPC increased overrun (start licking over 8.3 cm past the goal)
Supplementary Figure 2d; Results, Licking indicates action-dependent distance estimation, paragraph 3
How it will be judged
Per trial, the distance of the licking for reward acquisition relative to the goal (the first lick at or past the goal), PBS against muscimol trials per condition, Mann-Whitney U. Reproduced when muscimol trials lick farther past the goal with P < 0.01 in continuous and intermittent 1 (the legend's threshold); intermittent 2 (reported P = 0.0221) is shown.
muscimol-locomotion
PPC inactivation lowered locomotion speed and raised anticipatory licking in the continuous condition, but changed neither in the intermittent conditions.
Inactivation of PPC decreased the locomotion speed of mice in the continuous condition (MannWhitney U-test, P = 9.63 × 10−12, z = 6.81), which was most probably caused by increased anticipatory licking (Mann-Whitney U-test, P = 6.92 × 10−11, z = 6.52). In intermittent conditions, inactivation of PPC did not change the locomotion speed (P = 0.106, z = 1.62) and the licking (P = 0.0828, z = 1.73).
Online Methods, Muscimol injection, paragraph 2
How it will be judged
Per trial, muscimol against PBS trials, Mann-Whitney U (two-sided, Statistics); intermittent 1 and 2 pooled, as the text reports one test. The paper does not define the measures. Readings for speed, (a) mean speed from start to goal, (b) mean speed from start to reward. Readings for licking, (a) lick count from start to goal, (b) lick count from start to reward. The no-change parts are claims of absence, judged at P < 0.05. Reproduced when, under every reading, continuous speed is lower and licking higher after muscimol with P < 0.05, and neither differs with P < 0.05 in intermittent trials.
sound-responsive-layer3
PPC and PM neurons responded to sounds; in layer 3, PM had more sound-responsive neurons than PPC (4.78% against 1.48%).
On average, PM had more sound-responsive neurons than PPC in layer 3 (PPC, 1.48%, PM, 4.78%; Mann-Whitney U test, P = 0.0402)
Results, Goal-distance representation in PPC and PM, paragraph 2; Supplementary Figure 4b
How it will be judged
Sound-mapping trials (Online Methods, Imaging sound responses of neurons). Activity after sound from two adjacent speakers pooled into six azimuths; 500 ms moving average; a neuron is sound responsive when, at some azimuth, activity in the first 1.5 s after sound onset exceeds activity in the 1 s before (one-sided t-test, P < 1e-5). Percentages per field of view, readings of O5. Reproduced when, under one reading, both percentages are within tolerance and the per-session Mann-Whitney U test gives PM higher with P < 0.05.
neuron-classes
Goal-distance neurons numbered 1,116, 1,054 and 483 in PPC layers 2, 3 and 5 and 601, 479 and 129 in PM; of these, 744, 854 and 367 in PPC and 193, 204 and 54 in PM were after-reaching neurons; per session, on average 65.6, 70.3 and 34.5 in PPC and 37.6, 31.9 and 7.59 in PM were goal-distance neurons.
Number of after-reaching neurons in layers 2, 3, 5 was 744, 854, 367 in PPC and 193, 204, 54 in PM ... (average number of goal-distance neurons in layers 2, 3 and 5; PPC, 65.6, 70.3 and 34.5; PM, 37.6, 31.9 and 7.59)
Figure 2d; Figure 3c legend; Supplementary Figure 6 legend; Results, Distance prediction with dynamic Bayesian inference, paragraph 2
How it will be judged
The classification of the opening comment (Online Methods, Task-relevant neurons), under readings O1, O2 and O3. Before-reaching counts follow by subtraction and match Supplementary Fig. 9a (372, 200, 116; 408, 275, 75); the per-session averages are the totals over session counts. Fig. 2d also gives goal-distance or reward neurons (1,745, 1,456 and 651 in PPC; 816, 625 and 183 in PM), so reward neurons are 629, 402, 168; 215, 146, 54. These are shown, not judged. This claim selects the neurons other claims use. Reproduced when, under every reading, at least 10 of the 12 counts are within tolerance; partially reproduced when at least 6 are. The counts under each reading are shown.
fig2d-task-relevant-proportions
Task-relevant neurons made up 41.3, 33.8 and 20.0% of active neurons in PPC layers 2, 3 and 5 and 19.6, 16.7 and 5.70% in PM, more in PPC than in PM in every layer.
Among active neurons (99.6% of all neurons; Supplementary Table 1), the proportions of task-relevant neurons in layers 2, 3 and 5 were 41.3, 33.8 and 20.0% in PPC, and 19.6, 16.7 and 5.70% in PM, indicating that more neurons were recruited to the task in PPC than PM
Results, Goal-distance representation in PPC and PM, paragraph 4; Figure 2d
How it will be judged
Task relevance as in neuron-classes (O1, O2). Percentages under the two readings of O5 (Fig. 2d shows pooled counts 3,497, 2,376, 1,045 and 1,533, 1,180, 317 task-relevant neurons, which over the active counts give 41.8, 33.4, 21.5, 19.7, 16.8 and 5.37%. These differ from the reported values by up to 1.5 percentage points, so the reported values are probably per-session means, reading (a); the pooled values still lie within tolerance). Mann-Whitney U over per-session proportions, PPC against PM per layer. Reproduced when, under one O5 reading, all six values are within tolerance and, under every method reading, PPC exceeds PM with P < 0.05 in all three layers.
fig2d-goal-distance-ppc-more
PPC had more goal-distance neurons than PM in all three layers.
PPC had more goal-distance neurons than PM in all layers (Mann-Whitney U test, P = 0.00415, 0.0152 and 4.98 × 10−4, z = 2.87, 2.43 and 3.48, respectively; Fig. 2d)
Results, Goal-distance representation in PPC and PM, paragraph 4; Figure 2d
How it will be judged
Per session, goal-distance neurons, PPC against PM per layer, Mann-Whitney U over sessions. Whether counts or proportions of active neurons are compared is not said; readings (a) proportion of active neurons, (b) count. Reproduced when, under every reading of O1 to O3 and of the measure, PPC is higher with P < 0.05 in all three layers.
fig2d-after-vs-before
PPC had more after-reaching than before-reaching neurons in every layer, whereas PM tended the other way, significantly in layer 2.
PPC had more after-reaching neurons than before-reaching neurons (Wilcoxon signed-rank test, P = 0.0319, 0.00568 and 0.0176, z = 2.15, w = 13.5 and 9.5), whereas PM had the opposite tendency (P = 0.00771, 0.287 and 0.427, z = 2.66, w = 70 and 57.5)
Results, Goal-distance representation in PPC and PM, paragraph 4; Figure 2d
How it will be judged
Per session, counts (or proportions, O5) of after-reaching and before-reaching neurons, paired over sessions, two-sided signed-rank test per area and layer. Reproduced when, under every reading of O1 to O3, after-reaching exceeds before-reaching in all PPC layers with P < 0.05, and in PM before-reaching exceeds after-reaching in layer 2 with P < 0.05 and PM layers 3 and 5 do not show PPC's direction significantly.
fig2d-layer-distance-preference
In both PPC and PM, goal-distance neurons in layers 3 and 5 preferred distances closer to the goal than those in layer 2.
in both areas, goal-distance neurons in layers 3 and 5 were more tuned to the distance near the goal than those in layer 2 (Mann-Whitney U test, P = 0.00319 to 2.25 × 10−16)
Figure 2d insets and legend; Results, Goal-distance representation in PPC and PM, paragraph 4
How it will be judged
Distance preference per goal-distance neuron (O4), Mann-Whitney U over neurons between layers within each area, Bonferroni threshold P < 0.01 (Statistics, Fig. 2d). Reproduced when, under every reading of O1 to O3, layers 3 and 5 each have smaller preferred distances than layer 2 with P < 0.01 in both areas. The layer 3 against 5 comparisons are shown.
fig3b-distance-over-time
In PPC and PM, more goal-distance neurons were best fit by goal distance than by the time to reach the goal.
More neurons represented the goal distance than the time to goal in PPC and PM (Wilcoxon signed-rank test, P = 0.00348 to 4.33 × 10−4)
Figure 3b; Results, Goal-distance representation in PPC and PM, paragraph 6
How it will be judged
Single-neuron regression with BIC selection over the 11 variable sets (opening comment; O6). A neuron represents distance when its best set contains distance, and time when it contains time. Per session, counts of distance and of time neurons among goal-distance neurons, paired two-sided signed-rank test over sessions per area and layer. Reproduced when, under every reading, distance neurons outnumber time neurons with P < 0.01 in all six area-layer groups.
supp6-distance-not-sound
Goal-distance and after-reaching neurons were fit better by goal distance than by a distance held constant during no-sound zones or by the sound intensity.
RMSE of Distance was significantly smaller than of other variables both in goal-distance neurons and after-reaching neurons, indicating goal distance representation
Supplementary Figure 6; Results, Goal-distance representation in PPC and PM, paragraph 6
How it will be judged
Per neuron, y(t) = b0 + b1 phi(variable(t)) with the Gaussian basis of equation (3), variable Distance, Step (goal distance held at its value on entering each no-sound zone) or Sound (intensity, 100 dB SPL at 25 cm and 6 dB lower per doubling of distance; Online Methods). Intensity during no-sound zones and below 25 cm is not given; readings (a) zero during no-sound zones and the formula below 25 cm with distance floored at 0.42 cm; (b) the same with intensity capped at 100 dB below 25 cm. Frames as O6. RMSE of each fit, paired signed-rank test over neurons, Distance against Step and against Sound, for goal-distance and for after-reaching neurons in each area and layer. Reproduced when, under every reading, Distance has the lower RMSE with P < 0.05 in all 24 comparisons; partially when in at least 20.
tuning-similar-ppc-pm
The distance tuning curves of goal-distance neurons, as fitted coefficients, did not differ between PPC and PM.
Fitted coefficients for goal-distance representation revealed that the distance tuning curves of neurons were similar between PPC and PM (Online Methods; Mann-Whitney U test, P = 0.268 to 0.106, z = 1.11 to 1.62)
Results, Goal-distance representation in PPC and PM, paragraph 6
How it will be judged
A claim of absence. The coefficient compared is not named; the fit is the distance model of equation (2) for goal-distance neurons. Readings (a) the width s of the distance basis; (b) the distance coefficient b1. Mann-Whitney U over neurons, PPC against PM per layer, two-sided. Reproduced when no layer differs with P < 0.05; not reproduced when any does; effect sizes are shown. If the readings disagree the claim is blocked as methods_underspecified.
fig3c-continuous-ramp
In the continuous condition, after-reaching neurons increased their activity as mice approached the goal, in both PPC and PM.
Activity in the continuous condition gradually increased in both PPC and PM as mice approached the goal (Wilcoxon signed-rank test, P = 0.00507 to 0, z = 2.80 to 25.0; Fig. 3c, and Supplementary Figs. 7 and 8a)
Figure 3c; Results, PPC represents goal distance estimated from action, paragraph 1
How it will be judged
Per after-reaching neuron, its trial-averaged dR/R in the condition, binned every 0.42 cm (equation 5); whether the 8.37 cm moving average of the classification is applied first is not said, readings (a) not applied, (b) applied. Slope of activity against approach in the zones 25.1-8.4 and 8.4-0 cm (Supplementary Fig. 8a), signed-rank test of slopes against zero over neurons per area and layer. Reproduced when, under every reading, the median slope is positive (activity rising toward the goal) with P < 0.01 in all 12 area, layer and zone tests.
fig3c-ppc-ramp-without-sound
In the intermittent conditions, PPC after-reaching neurons in all layers increased activity toward the goal both with and without sound.
In intermittent conditions, increased activity was observed with and without sound inputs in all PPC layers (sound zones ... P = 1.37 × 10−4 to 0, z = 3.81 to 23.2; no-sound zones ... P = 1.19 × 10−6 to 0, z = 4.86 to 17.9)
Figure 3c; Supplementary Figure 8a; Results, PPC represents goal distance estimated from action, paragraph 1
How it will be judged
Slopes as in fig3c-continuous-ramp (both readings), in intermittent 1 and 2 trials, zones 25.1-8.4 and 8.4-0 cm (one with and one without sound in each condition). Reproduced when, under every reading, the median slope is positive with P < 0.01 in all 12 PPC tests (3 layers, 2 conditions, 2 zones).
supp8a-pm-no-sound
In PM, after-reaching neurons showed positive slopes in some no-sound zones, but in the intermittent 1 condition their slopes without sound were not significantly positive.
Slopes were significantly positive with and without sound in PPC in all conditions, but in PM in intermittent 1 condition they were not significantly positive without sound.
Supplementary Figure 8a; Results, PPC represents goal distance estimated from action, paragraph 1
How it will be judged
Slopes as in fig3c-continuous-ramp, PM after-reaching neurons, no-sound zones (intermittent 1, 8.4-0 cm; intermittent 2, 25.1-8.4 cm), signed-rank against zero per layer, P < 0.01 (the legend's threshold). Supplementary Fig. 8a marks all three PM layers significant in intermittent 2 at 25.1-8.4 cm and none in intermittent 1 at 8.4-0 cm. Reproduced when, under every reading, all three PM intermittent 2 no-sound tests are positive with P < 0.01 and no PM intermittent 1 no-sound test is (the latter a claim of absence, judged by the paper's test). Per layer results are shown.
supp8b-slopes-lower-without-sound
Slopes of after-reaching activity were lower without sound than with sound.
Although slopes were significantly positive even without sound in a, they were less steep than those with sound
Supplementary Figure 8b
How it will be judged
For each after-reaching neuron and zone (25.1-8.4 and 8.4-0 cm), the slope with sound (from the condition with sound in that zone) against without (the other intermittent condition), paired signed-rank over neurons. Reproduced when, under every reading of fig3c-continuous-ramp, the slope without sound is lower with P < 0.01 in all six PPC comparisons and in at least four of six PM comparisons (the paper reports five).
supp8c-ppc-steeper-than-pm
Without sound, PPC after-reaching neurons in layers 2 and 3 had steeper slopes than PM, while slopes in PM were equal to or lower than in PPC.
In layers 2 and 3, PPC had steeper slopes than PM without sound
Supplementary Figure 8c; Results, PPC represents goal distance estimated from action, paragraph 1
How it will be judged
No-sound slopes per neuron as in supp8b-slopes-lower-without-sound, PPC against PM neurons per layer and zone, Mann-Whitney U. The legend gives two P values per zone and layer; they are read in the order intermittent 1, intermittent 2, as the other legends list conditions, which puts the without-sound value second in 25.1-8.4 cm and first in 8.4-0 cm (the only order under which the legend supports the text's layers 2 and 3). The figure confirms it, since in each zone the red intermittent 1 pair is drawn first, and the stars fall on the no-sound pairs as read. Reproduced when, under every reading, PPC no-sound slopes exceed PM with P < 0.05 in layer 3 in both zones and in layer 2 in at least one zone, and PM is significantly steeper in no comparison.
supp9-preserved-representation
PPC neurons kept the distance preference they had in the continuous condition in the intermittent conditions, especially near the goal, more than PM neurons did.
PPC neurons preserved the distance representation, especially near the goal, significantly more than neurons in PM (Mann-Whitney U test, P = 0.0429 to 6.82 × 10−14, z = 2.02 to 7.49)
Supplementary Figure 9b; Results, PPC represents goal distance estimated from action, paragraph 2
How it will be judged
Per goal-distance neuron, the difference between its distance preference (O4) in continuous trials and in intermittent 1 (and 2) trials. The legend says only "difference"; it is read as the absolute difference, the only reading under which a smaller value means a better preserved preference (signed differences are shown); PPC against PM per layer and neuron group, Mann-Whitney U. Groups as Supplementary Fig. 9a, before-reaching neurons in four groups by continuous preference and after-reaching neurons; the four groups' borders are not given, readings (a) equal 16.75 cm quarters of 67 to 0 cm, (b) the zone borders 58.6, 25.1 and 8.4 cm. Reproduced when, under every reading, in intermittent 1 PPC after-reaching neurons have the smaller difference with P < 0.05 in at least one layer, and PM has a significantly smaller difference in no comparison. "Especially near the goal" is read as the after-reaching group, the one nearest the goal.
fig4-above-chance
Decoders built from goal-distance neurons estimated goal distance above chance in both PPC and PM.
Both areas estimated the goal distances reliably above chance level (Wilcoxon signed rank test, P = 0, z = 181 to 439)
Results, Distance prediction with dynamic Bayesian inference, paragraph 2; Figure 4b
How it will be judged
Decoder of the opening comment (O7, O8). Chance level is not defined. The z values imply a paired test over frames. Readings of chance (a) the RMSE of the flat prior over 0 to 67 cm at each frame; (b) the RMSE of a decoder whose training trials have their distances permuted across trials. Paired signed-rank test of decoder RMSE against chance over frames, per area, layer and condition. Reproduced when, under every reading, decoder RMSE is lower with P < 0.01 in all 18 settings.
fig4c-ppc-lower-error
The decoding error (RMSE) from layer 2 neurons was lower in PPC than in PM.
the estimation in PPC had significantly less deviation from the actual distance than in PM (Mann-Whitney ... U-test in layer 2, P = 0.00765 to 0, z = 2.67 to 17.1)
Figure 4c; Results, Distance prediction with dynamic Bayesian inference, paragraph 2
How it will be judged
Per trial, mean RMSE within each goal-distance bin (O8), PPC trials against PM trials per bin and condition, Mann-Whitney U, P < 0.01 (legend threshold). The legend's ranges (continuous P = 0.893 to 0) show that not every bin differs. Reproduced when, under every reading, in each condition PPC has the lower median RMSE in more than half of the bins and significantly lower RMSE in at least one, and PM is significantly lower in no bin. Layers 3 and 5 (Supplementary Fig. 10) are shown.
fig4d-map-near-goal
PPC and PM gave similar MAP estimates, but near the goal PPC estimates were better than PM.
PPC and PM had similar MAP estimates, but the estimations near the goal were better in PPC than in PM (P = 0.00549 to 2.47× 10−11, z = 2.78 to 6.68; Fig. 4d)
Figure 4d; Results, Distance prediction with dynamic Bayesian inference, paragraph 2
How it will be judged
Per trial, mean MAP within each bin (O8), PPC against PM, Mann-Whitney U per bin, layer 2, P < 0.01. "Near the goal" is read as the bins between 8.4 and 0 cm. Reproduced when, under every reading, in the continuous condition at least one bin in 8.4-0 cm has PPC MAP closer to the actual distance with P < 0.01 and no bin in that range has PM closer with P < 0.01.
fig4e-ppc-lower-uncertainty
The uncertainty (posterior s.d.) of distance estimates was lower in PPC than in PM at every goal distance and in every condition.
uncertainty of distance estimation was significantly less in PPC than PM (P = 1.09 × 10−12 to 0, z = 7.12 to 22.9; Fig. 4e)
Figure 4e; Results, Distance prediction with dynamic Bayesian inference, paragraph 2
How it will be judged
Per trial, mean posterior s.d. within each bin (O8), PPC against PM per bin and condition, layer 2, Mann-Whitney U. Reproduced when, under every reading, PPC is lower with P < 0.01 in every bin of every condition; partially when in at least 90% of bins. Layers 3 and 5 (Supplementary Fig. 10) are shown.
supp11-sound-reduces-error
Decoding errors decreased with sound and increased without it. Within zones, the change per cm of the RMSE difference (intermittent minus continuous) was lower in sound zones than in no-sound zones, in all PPC layers and in PM layers 2 and 3.
Decoding errors decreased with sound and increased without sound (Mann-Whitney U test, P = 0.00295 to 1.56 × 10−6, z = 2.97 to 4.80; Supplementary Fig. 11)
Supplementary Figure 11b,c; Results, Distance prediction with dynamic Bayesian inference, paragraph 2
How it will be judged
Supplementary Fig. 11b,c. Per session, dRMSE, the difference of median RMSE (over trials) between each intermittent condition and the continuous condition in 0.42-cm bins. Linear regression of dRMSE on distance within each zone, y = b0 - b1 distance, so b1 is the rise per cm of approach (Fig. 11c axis, "dRMSE per cm"). Then b1 of sound zones against b1 of no-sound zones, Mann-Whitney U. Fig. 11b shows four zones per intermittent condition (58.6-41.9, 41.9-25.1, 25.1-8.4 and 8.4-0 cm). The legend's "n = 68 (that is, 17 sessions times 4)" allows two readings. (a) Four sound and four no-sound zones per session, n = 68 per group. (b) 68 zones in all, the zones 25.1-8.4 and 8.4-0 cm of both conditions, two sound and two no-sound per session. Decoder sessions exclude the 3 with no goal-distance neurons, which fits PM layer 5's 14 sessions. Reproduced when, under every reading, b1 is lower with sound with P < 0.01 in all PPC layers. PM is shown.
fig5a-map-follows-continuous
In the continuous condition, decoded MAP distances followed the actual distance in all layers of PPC and PM.
In the continuous condition, MAPs followed the actual distances in all layers of PPC and PM (Wilcoxon signed-rank test, P = 9.56 × 10−14 to 0, z = 7.45 to 34.1; Fig. 5a)
Figure 5a; Results, Distance prediction with dynamic Bayesian inference, paragraph 3
How it will be judged
Per trial and zone (O8, four zones), MAP decrease per cm (equation 10); signed-rank test of slopes against zero over trials. Reproduced when, under every reading, the median slope is positive with P < 0.01 in all 24 area, layer and zone tests.
fig5a-map-decreases-without-sound
In the intermittent conditions, MAPs decreased with and without sound in most zones, and in every zone in PPC layers 3 and 5.
In intermittent conditions, MAPs decreased with and without sound inputs in most distance zones, and especially in all zones in PPC layers 3 and 5 (sound zones ... P = 0.00293 to 0, z = 2.98 to 14.7; no-sound zones ... P = 6.59 × 10−4 to 0, z = 3.41 to 8.56)
Figure 5a; Results, Distance prediction with dynamic Bayesian inference, paragraph 3; Discussion, paragraph 3
How it will be judged
As fig5a-map-follows-continuous, in intermittent 1 and 2 trials. Reproduced when, under every reading, in PPC layers 3 and 5 every zone of both conditions (16 tests) has a positive median slope with P < 0.01, and over all 48 tests more than half do.
fig5b-map-slope-lower-without-sound
MAP slopes were shallower without sound than with sound.
Slopes of MAPs without sound were lower than those with sound (Mann-Whitney U test, P = 5.14 × 10−12 to 0; Fig. 5b)
Figure 5b; Results, Distance prediction with dynamic Bayesian inference, paragraph 3
How it will be judged
Per trial MAP slopes in the zones 25.1-8.4 and 8.4-0 cm; for each zone, trials of the condition with sound there against trials of the one without, Mann-Whitney U (Fig. 5b legend). Reproduced when, under every reading, the no-sound slope is lower with P < 0.01 in all six PPC comparisons and in at least four of six PM comparisons.
fig5-layer3-steeper-than-layer2
In intermittent 1, without sound between 8.4 and 0 cm, MAPs decreased more steeply in PPC layer 3 than layer 2, while the s.d. of the estimates did not differ between these layers.
In the intermittent 1 condition, the decrease of MAPs in PPC layer 3 was significantly steeper than that in layer 2 without sound between 8.4 and 0 cm (Mann-Whitney U test, P = 0.00685, z = 2.70), whereas the s.d. of distance estimation showed no significant difference (P = 0.247, z = 1.16)
Results, Distance prediction with dynamic Bayesian inference, paragraph 3
How it will be judged
Per trial MAP slope and s.d. slope (equation 10) in intermittent 1 trials, zone 8.4-0 cm, PPC layer 3 trials against layer 2 trials, Mann-Whitney U. The s.d. part is a claim of absence judged at P < 0.05. Reproduced when, under every reading, the layer 3 MAP slope is steeper with P < 0.05 and the s.d. slopes do not differ with P < 0.05.
supp13-map-speed
During no-sound zones, decoded MAPs fell faster in fast-locomotion trials than in slow ones.
Slopes of MAPs were steeper in fast trials than slow trials, suggesting action-dependent distance prediction of mice
Supplementary Figure 13; Results, Distance prediction with dynamic Bayesian inference, paragraph 3
How it will be judged
Per trial, MAP regressed on time within no-sound zones (y(t) = b0 + b1 t); trials split into slow and fast halves (O9); fast against slow trials per area, layer and condition, Mann-Whitney U. Reproduced when, under every reading, fast trials have the steeper median slope with P < 0.05 in PPC layers 2 and 3 in both conditions; the other eight tests are shown (the paper reports PM significant in intermittent 1 only).
fig5cd-uncertainty-updating
In all PPC layers, between 25.1 and 0 cm, uncertainty of the estimated distance fell more steeply with sound than without, the updating step of dynamic Bayesian inference.
Uncertainty of estimated distance decreased with sound inputs compared with no sound in all layers of PPC between 25.8 and 0 cm (P = 0.0153 to 6.92 × 10−11; Fig. 5c,d)
Figure 5c,d; Results, Distance prediction with dynamic Bayesian inference, paragraph 3; Discussion, paragraph 3
How it will be judged
Per trial s.d.-of-posterior slopes (decrease per cm, equation 10) in the zones 25.1-8.4 and 8.4-0 cm, sound trials against no-sound trials per zone as in fig5b-map-slope-lower-without-sound, Mann-Whitney U. Reproduced when, under every reading, the decrease is larger with sound with P < 0.05 in all six PPC comparisons. PM results are shown (the paper reports them mixed).
fig5c-uncertainty-decreases-with-sound
In all PPC layers, the uncertainty of the decoded distance (s.d. of the posterior) decreased during sound presentation between 8.4 and 0 cm.
Uncertainty decreased during sound presentations in PPC between 8.4 cm and 0 cm.
Figure 5c and legend
How it will be judged
Per trial s.d.-of-posterior slope (decrease per cm, equation 10) in the zone 8.4-0 cm, signed-rank test against zero over trials per area, layer and condition, P < 0.01 (Fig. 5c, "Data presentation as in a"). The legend gives only the range over all Fig. 5c tests (P = 0.892 to 0), so no statistic is recorded. Fig. 5c marks, in the 8.4-0 cm zone of each PPC layer, the continuous and intermittent 2 tests, the two conditions with sound there. Reproduced when, under every reading of O7 and O8, the decrease is positive with P < 0.01 in all six (three PPC layers, continuous and intermittent 2). PM and intermittent 1 are shown.
lognormal-decoder-better
The decoder with a log-normal likelihood estimated goal distance better than one with a Gaussian likelihood in 17 of 18 settings, all but PM layer 5 in the continuous condition.
The decoder with log-normal distribution gave better estimates for goal distance than that with a Gaussian distribution in the 18 settings ... except for layer 5 in PM in continuous condition (P = 0.834, z = 0.209)
Online Methods, Probabilistic decoding of goal distance, paragraph 3
How it will be judged
The same decoder with a Gaussian likelihood of raw activity (mean and s.d. per bin as for the log-normal). "Better" is not defined; readings (a) lower RMSE, (b) lower absolute MAP error, paired over frames (the z values imply frames), signed-rank test per setting (3 layers x 2 areas x 3 conditions). Reproduced when, under every reading, the log-normal decoder is better with P < 0.01 in at least 16 of the 18 settings.
svm-correlations
An SVM decoder trained on recorded data did not estimate distance better than one trained on trial-shuffled data, except in one setting, so ignoring correlations between neurons cost the decoder little.
Recorded-SVM did not estimate distance better than shuffled-SVM (one-sided Wilcoxon signed-rank test, P = 0.957 to 1, z = 1.71 to 12.5), except in 1 setting (layer 3 in PM in intermittent1 condition, P = 0.00572, z = 2.53)
Online Methods, Probabilistic decoding of goal distance, paragraph 8
How it will be judged
A claim of absence. SVM (Matlab) on goal-distance neurons' activity, trained and validated as the probabilistic decoder. The paper says only "shuffled data" (refs. 49, 50); the shuffle is read as the standard one of those references, each neuron's activity permuted independently across training trials at the same goal distance (4.2-cm bin). Kernel, classes and error measure are not given; readings (a) linear multiclass SVM over 4.2-cm distance bins, error the absolute difference of the predicted bin centre from the actual distance; (b) linear SVM regression, absolute error. One-sided signed-rank test (recorded better) over frames per setting. Reproduced when, under every reading, no more than one of the 18 settings shows the recorded SVM better with P < 0.01; not reproduced when more do.

Pipeline

Paper: done (1 of 1)
Claims: done
Claims check: done (ready)
Attempt: todo
Questions: none
Verified: none