Claims 27
From the paper High-Density Recording Reveals Sparse Clusters (But Not Columns) for Shape and Texture Encoding in Macaque V4 · claims file · checked 2026-10-21: ready
- methods-sessions-and-neurons
The shape test was run in 30 sessions in M1 and 44 in M2, the texture test in 17 and 34, and the position control in 33 M2 sessions; the responsiveness criterion excluded 461 and 366 neurons and left 1,850 neurons for shape and 979 for texture analyses.
We conducted 30 and 44 sessions of the main shape test in M1 and M2, respectively. The position control experiment with shape stimuli was conducted in 33 of 44 sessions in M2 only. The texture test was conducted in 17 and 34 sessions in M1 and M2, respectively ... This excluded 461 and 366 neurons across our dataset for shape and texture analyses and left 1,850 and 979 neurons, respectively
Materials and Methods, Dataset and data inclusion criteria, paragraphs 1 and 2How it will be judged
Sessions counted per monkey and test. Responsive neurons are good units passing the 5 spikes/s criterion of the opening comment, counted once per session and test; excluded neurons are good units of those sessions that fail it. Reproduced when every value is within tolerance; partially reproduced when the session counts are and a neuron count is not.
- ex-m2p6-cluster
In penetration M2p6, 24 well-isolated neurons were recorded over about 2,400 um; a group of superficial neurons (1 to 9, over 540 um) had similar shape tuning at both contrasts, and with dark shapes neurons 1 to 9, 14 and 17 had a significant positive Isimilarity with a negative slope, while most other, deeper neurons did not.
We recorded 24 well-isolated neurons in this session (Fig. 2A, left, #1-#24; superficial to deep) over a ∼2,400 μm length of the probe ... a small group of neurons (#1-#9) spread over a 540 μm extent of the probe in the superficial cortex (Fig. 2A, left) exhibited high similarity in their responses to the tested shapes ... For some neurons (Neurons #1-9, #14, and #17), regression lines were characterized by a large positive y-intercept (Isimilarity) significantly different from zero (p < 0.05, solid lines) and a negative slope (Ssimilarity)
Results, Example penetration, paragraphs 1 and 2; Figures 2A, 3A and 3CHow it will be judged
M2p6 is not identified in the paper beyond its name; the readings are (a) the sixth M2 session in recording order among shape sessions, (b) the sixth M2 session in recording order among all sessions, (c) the M2 session the dataset itself labels p6 or penetration 6, each accepted only when its responsive-neuron count is within tolerance of 24. Neurons are numbered by depth (Figure 2A legend). Isimilarity with dark shapes as in the opening comment. Reproduced when the expected values hold, the neurons with a significant positive intercept are neurons 1 to 9, 14 and 17 with at most one neuron added or missing (the count's tolerance), and their slopes are negative; partially reproduced when the session is found and the set or the count does not match.
- ex-m2p6-pairs
In M2p6, neurons 3 and 8, 260 um apart, have tuning similarity 0.82 at both contrasts, while neurons 18 and 19, 80 um apart, have similarity 0.00 (dark) and -0.25 (bright), though both pairs respond over a broad range.
For example, Neurons #3 and #8, recorded on contacts 260 μm apart, share a great deal of similarity in tuning (rsimilarity = 0.82), but Neurons #18 and #19 recorded on contacts 80 μm apart do not (rsimilarity= 0; Fig. 3B)
Results, Example penetration, paragraph 2; Figure 3BHow it will be judged
Noise-corrected similarity as in the opening comment. The paper gives the distances between the contacts nearest each neuron's center of mass (Figure 2A legend); the template center-of-mass distance is shown as a variant. The tolerance of the zero value is 0.1, since 10% of zero is no tolerance; 0.1 is 5% of the measure's range of 2. Reproduced when every value is within tolerance; partially reproduced when the 3 and 8 values are and an 18 and 19 value is not, or the reverse.
- ex-m2p6-invariance
In M2p6, shape tuning was contrast-invariant (invariance above 0.5) for 19 of the 24 neurons, whatever their Isimilarity.
For a majority of neurons recorded on this day (19/24 neurons), contrast invariance of shape tuning was high (>0.5) regardless of the value of Isimilarity
Results, Example penetration, paragraph 3; Figure 3DHow it will be judged
Tuning invariance as in the opening comment, over the 120 shapes. Unreliable values count as not above 0.5 (the Figure 4F legend plots an unreliable invariance at 0). Reproduced when the count is within tolerance under reliability reading (a) or (b).
- ex-csd-sinks
The CSD shows a current sink taken as layer 4 at about 1,073 um below the most superficial neuron, 53 ms after stimulus onset, in M2p6, and at 570 um, 52 ms, in M1p12, with no other sink deeper in M1p12.
Current source density analysis (CSD, see Materials and Methods) reveals a current sink at ∼53 ms after stimulus onset at a recording depth of ∼1,073 μm relative to the most superficially recorded neuron in this session ... We found that a current sink with short latency (∼52 ms) emerged at a shallow depth along the penetration (570 of 2,580 μm, Fig. 4G), but no other sinks were evident deeper in the cortex
Results, Example penetration, paragraph 1, and paragraph 4; Figures 2B and 4GHow it will be judged
CSD as in Materials and Methods, Current source and sink profile along the probe. LF downsampled to 1 kHz, mean subtracted, scaled to volts, third-order Butterworth band-pass 0.5 to 100 Hz; the LF after the first stimulus of each trial averaged over trials, Gaussian smoothed over 10 adjacent contacts, second spatial derivative over three neighboring contacts, Gaussian smoothed over 10 contacts again, sign inverted, per column of contacts. OPEN, the Gaussian's width, the column shown, and how the sink is located are not given; the readings are (a) the earliest sink peak after stimulus onset and (b) the largest sink in 0 to 100 ms, each on every column, reproduced when one column under one reading gives all values within tolerance. Depth is relative to the most superficial good unit. Needs the LF signal. OPEN, "no other sinks were evident deeper" in M1p12 has no criterion in the paper (it gives no rule for what counts as a sink); that part is shown as the CSD profile and not judged.
- ex-m1p12-no-cluster
In penetration M1p12, 42 neurons over 2,580 um showed no clustered shape tuning at either contrast; only one neuron had a significant Isimilarity, although most had high contrast invariance of shape tuning.
Here, we studied the responses of 42 neurons over a 2,580 μm distance (Fig. 4A) ... The y-intercept for the regression line was significantly different from zero for only one neuron (solid line) ... Neurons in this penetration did not exhibit shape tuning that was similar to their neighbors (filled), but most neurons did exhibit high shape tuning invariance
Results, Example penetration, paragraph 4; Figures 4A, 4B, 4D and 4FHow it will be judged
M1p12 is found as M2p6 is in ex-m2p6-cluster (readings a to c, accepted when the count is within tolerance of 42). Isimilarity with dark shapes, as Figure 4D is said to follow Figure 3C. The count of 1 takes an absolute tolerance of 1, as 10% of 1 is less than one neuron; reproduced when at most 2 neurons have a significant intercept and more than half of the neurons have reliable tuning invariance above 0.5 (the threshold the paper uses for "high" in the M2p6 example); partially reproduced when the count holds and the invariance part does not.
- ex-m1p12-strong-responses
In M1p12, peak responses exceeded 15 spikes/s in 28 neurons for dark and 26 for bright shapes, so the lack of clustering was not due to weak responses.
peak responses were often high (exceeding 15 spikes/s in 28/26 neurons for dark/bright stimuli)
Results, Example penetration, paragraph 4; Figure 4CHow it will be judged
Peak response is the largest mean rate over the 120 shapes of one contrast, 30 to 200 ms. OPEN whether baseline is subtracted; Figure 4C plots baseline-subtracted rates. The readings are (a) baseline-subtracted and (b) raw. Reproduced when both values are within tolerance under one reading.
- pop-significant-isimilarity
Of 1,850 neurons in 74 sessions (6 to 76 per penetration), a minority, 344 (203 in M1, 141 in M2), had a significantly positive Isimilarity with a negative Ssimilarity, reflecting similarity with nearby neurons that declined with distance; the majority (1,504) had an Isimilarity not significantly above zero.
We studied the responses of 1,850 neurons across 74 recording sessions (M1, 30; M2, 44), ranging from 6 to 76 neurons per penetration ... we found a minority of neurons (344/1,850; M1, 203; M2, 141; Fig. 5A, blue) with positive Isimilarity, significantly different from zero (p < 0.05), and a negative Ssimilarity ... The majority (n = 1,504) had Isimilarity that was not significantly greater than zero
Results, Population results, shape tuning, paragraph 1; Figure 5AHow it will be judged
Isimilarity as in the opening comment, under contrast readings (a) dark, (b) bright, (c) either, crossed with reliability readings (a) and (b). OPEN whether the 344 also require a negative slope; 344 + 1,504 = 1,848, the neurons with a line, suggests the count is of significant positive intercepts alone, the slope being described. The readings are (i) significant positive intercept and (ii) significant positive intercept with negative slope. Reproduced when one reading puts the four counts within tolerance.
- pop-intercept-slope-correlation
Across neurons Isimilarity and Ssimilarity are strongly negatively correlated (r = -0.745), and a significantly positive slope is almost absent; only one neuron (in M2) had both a significantly positive Isimilarity and a significantly positive Ssimilarity.
Overall, Isimilarity and Ssimilarity were significantly negatively correlated (r = −0.745, p < 0.0001, n = 1,848), and we seldom found neurons associated with positive Ssimilarity, i.e., similarity that increased with distance: only one neuron (in M2) was associated with a significantly positive Isimilarity and Ssimilarity
Results, Population results, shape tuning, paragraph 1How it will be judged
OPEN which correlation; the paper names Pearson's for other plain r values (Figure 5E) and Spearman's where it uses it, so the readings are (a) Pearson and (b) Spearman, under the contrast readings of the opening comment. Reproduced when one reading gives r within tolerance with p < 0.0001 and at most 2 neurons with both significantly positive (the count of 1 takes an absolute margin of 1, as a relative tolerance on 1 is meaningless).
- pop-dark-bright-consistent
Isimilarity from dark and from bright shapes is consistent across neurons (Spearman's r = 0.435, n = 1,847).
We also found that tuning similarity (Isimilarity) based on responses to dark and bright stimuli was consistent (Fig. 5B)
Results, Population results, shape tuning, paragraph 1; Figure 5BHow it will be judged
Neurons with an Isimilarity at both contrasts (Materials and Methods, Statistical tests). Reproduced when r is within tolerance and p < 0.0001, under either reliability reading.
- pop-contrast-invariance
Contrast-invariant shape tuning is widespread; more than half of the neurons are well driven and contrast invariant, with a median correlation between dark and bright responses of 0.64.
contrast-invariant shape tuning was widely prevalent across our dataset ... more than half of the recorded neurons were well-driven and exhibited contrast-invariant shape tuning (median correlation between responses to dark and bright stimuli in individual penetrations = 0.64)
Results, Population results, shape tuning, paragraph 2How it will be judged
Tuning invariance as in the opening comment. The Methods (Statistical tests) take "the median of tuning invariance for shape for all recorded neurons"; the Results say "in individual penetrations". The readings are (a) the median over all neurons, (b) the median of the per-penetration medians, each with unreliable values (c) left out or (d) counted as 0. Reproduced when one reading gives a median within tolerance.
- pop-sparse-across-penetrations
Neurons with significant Isimilarity occur in small, sparse clusters in 55 of the 74 penetrations; only 28 penetrations have five or more such neurons, and 20 penetrations yielded more than 30 neurons.
Neurons with high similarity in tuning (Fig. 5A) were widely distributed across 55 of the 74 penetrations in small, sparse clusters (blue). Only 28 penetrations had five or more neurons with significant Isimilarity (Fig. 5A asterisks) ... simultaneous recordings from >30 neurons (achieved on 20/74 penetrations)
Results, Population results, shape tuning, paragraph 2; Figure 5AHow it will be judged
Under the readings of pop-significant-isimilarity. Reproduced when one reading puts all three values within tolerance.
- pop-asymptote-20-percent
When more than 30 neurons were recorded, the percentage of neurons with significant Isimilarity levels off at about 20%, so shape clusters are not columnar.
When the number of recorded neurons was >30, the percentage of neurons with significant Isimilarity asymptoted to ∼20% (Fig. 5C) suggesting that shape clusters were not "columnar" extending across laminae in V4
Results, Population results, shape tuning, paragraph 2; Figure 5CHow it will be judged
OPEN how the asymptote is summarized; the paper shows a scatter. The readings are (a) the median of per-penetration percentages over penetrations with more than 30 neurons and (b) the pooled percentage over their neurons, under the readings of pop-significant-isimilarity. Reproduced when one reading is within tolerance.
- pop-mean-isimilarity-yield
In penetrations with few neurons the mean Isimilarity varies widely across penetrations; with more than 30 neurons it levels off.
For penetrations with low neural yield, the standard deviation of Isimilarity across neurons was large leading to great variability in the mean across penetrations; with larger yield (>30 neurons), mean of Isimilarity asymptoted (Fig. 5D)
Results, Population results, shape tuning, paragraph 2; Figure 5DHow it will be judged
The paper reports no test. Per penetration the mean Isimilarity. Reproduced when the standard deviation of the penetration means is smaller for penetrations with more than 30 neurons than for those with 30 or fewer, under the contrast reading that reproduces pop-significant-isimilarity (all three shown); the two standard deviations and a Levene test are reported.
- pop-recording-density
The fraction of neurons with significant Isimilarity is only weakly related to the recording density (Pearson's r = 0.236, p = 0.0429).
We did not find a strong relationship between recording density and fraction of neurons with significant Isimilarity (Fig. 5E) ... Pearson's correlation coefficient, r = 0.236, p = 0.0429
Results, Population results, shape tuning, paragraph 2; Figure 5E legendHow it will be judged
The legend defines density as the depth range of the penetration's neurons divided by their number, while the axis reads number of neurons per um. The readings are (a) neurons per um and (b) um per neuron, crossed with the readings of pop-significant-isimilarity. Reproduced when one reading gives r within tolerance and p < 0.05.
- pop-within-vs-across-sessions
Neurons within a penetration have slightly more similar shape tuning than neurons from different sessions (mean correlation 0.0582 against 0.0022), and more so within 200 um (0.0918 against 0.0435 further apart), but all means are weak.
The mean of the correlation coefficient between the responses of neuron pairs recorded in different sessions was slightly but significantly less than that between simultaneously recorded neuron pairs (0.0022 vs 0.0582, respectively; Welch's t test, p < 0.0001) ... The mean correlation coefficient was even greater when it was based on neurons within a penetration separated by ≤200 μm (0.0918) than neurons that were further apart (0.0435; Welch's t test, p < 0.0001). Importantly, the means were quite weak in all three cases
Results, Population results, shape tuning, paragraph 3How it will be judged
Welch's t test as in Materials and Methods, Statistical tests. OPEN, all able to change the means. (1) The coefficient; "the correlation coefficient between shape responses" may be (a) Pearson's r of mean rates over the 120 shapes or (b) the noise-corrected similarity. (2) The contrast; (a) dark, (b) bright, (c) both contrasts as separate pairs. (3) The cross-session pairs; (a) all pairs of responsive neurons from different sessions or (b) only pairs within one monkey. Reproduced when both tests give p < 0.0001 in the stated direction under every reading and all four means are within tolerance under one reading; partially reproduced when the directions and p hold under every reading and no reading matches the means. The 10% tolerance on means near zero is very tight; the size of the differences is reported either way.
- orientation-sf-isimilarity
Tuning similarity computed from orientation x spatial frequency fits (12 Gabor filters) to the shape responses is also diverse among nearby neurons, and its Isimilarity correlates with the Isimilarity from the shape responses (Spearman's r = 0.588, n = 1,848).
Our results show that tuning preferences for orientation were quite diverse for nearby neurons, similar to our results based on shape responses (Fig. 6A). Across all penetrations, Isimilarity based on orientation × SF was strongly correlated with those based on shape responses
Results, Tuning for stimulus orientation, paragraph 1; Figures 6A and 6BHow it will be judged
Materials and Methods, Feature models; shape responses fit as a linear weighted sum of the outputs of 12 Gabor filters (4 orientations x 3 spatial frequencies) applied to the shape images; similarity is Pearson's r of fitted responses; Isimilarity as in the opening comment. Needs the shape images (Pasupathy and Connor 2001). OPEN, able to change the result; the filters' spatial frequencies, envelope, phase and pooling, and the fitting method, are not given (the paper cites Portilla and Simoncelli 2000 for the filters and David et al. 2006 for the orientation x SF model). The readings are (a) the energy of each subband of a steerable pyramid with 4 orientations and 3 scales summed over the image, (b) quadrature Gabor energy at 4 orientations and 3 octave-spaced frequencies summed over the image, and (c) the image's Fourier power pooled in 4 orientation x 3 SF bands, as in the spectral model of David et al. 2006, each fit by least squares; with the contrast readings of the opening comment. Reproduced when all readings give r within tolerance and p < 0.0001; blocked as methods_underspecified when they differ in outcome.
- orientation-auc-coarse-clusters
Assayed at a coarse scale with two orthogonal orientations of the spindle shape, some penetrations consist entirely of neurons preferring horizontal or vertical, and in others the preference switches along the probe, as with optical imaging, even where single-neuron shape clustering is not extensive.
We found penetrations that entirely consisted of neurons with preference for horizontal or vertical orientations (Fig. 6C) or penetrations where preference of neurons switched from one to the other (Fig. 6D), similar to what has been observed with optical imaging techniques
Results, Tuning for stimulus orientation, paragraph 2; Figures 6C and 6DHow it will be judged
Materials and Methods, Feature models; per neuron the AUC between spike counts (30 to 200 ms, per presentation) for shape 1 at 0 and 180 degrees and at 90 and 270 degrees, both contrasts pooled, above 0.5 meaning vertical; averaged over neurons in 100 um depth bins. The examples are M2p6 and M1p5 (Figure 6C, D). The paper defines neither kind of penetration; the plain readings of "entirely" and "switched" are used, not taken from the paper. A penetration is all-one-preference when every occupied bin is on one side of 0.5, and switching when occupied bins lie on both sides, each side in a contiguous run of depth. OPEN which rotation is horizontal depends on the dataset's rotation labels and on shape 1 being the spindle (Figure 1B). Reproduced when at least one penetration of each kind exists among the 74 and M2p6 and M1p5 are each of one kind; the counts of each kind are reported.
- apc-peaks-sparse-clustering
Along angular position and curvature separately, clustering of tuning peaks is sparse; of 59 penetrations with at least four neurons well fit by the APC model, only 7 (angular position) and 5 (curvature) show a significant positive correlation between peak difference and distance, and only 9 and 5 sessions have smaller mean and STD of the differences than a permutation null.
Across 59 penetrations which included at least four neurons well-fit by the angular position × curvature model, only 7 (5) penetrations exhibited statistically significant positive correlation between the ΔAP (ΔCV) and inter-neuron distance (data not shown) ... we found that a minority (nine and five) of sessions exhibited smaller mean and STD than the null distributions (data not shown). These results indicate that evidence for clustered tuning ... is sparse in our dataset
Results, Tuning for curvature or angular position, paragraph 1How it will be judged
Materials and Methods, Feature models and Statistical tests; shape responses fit with the 4D APC Gaussian of Pasupathy and Connor (2001) (one angular position Gaussian, three curvature Gaussians for adjoining segments); well fit means goodness of fit r > 0.5 (Figure 7A legend); delta AP is the circular absolute difference of AP peaks (0 to 180 degrees), delta CV the absolute difference of central curvature peaks; per penetration Pearson's r of the difference against inter-neuron distance over all pairs; the permutation null shuffles AP and CV values across penetrations, 1,000 times, and a session counts when its mean and STD are below the one-tailed 95% limit. Needs the shapes' boundary curvature descriptions (Pasupathy and Connor 2001). OPEN, able to change counts; (1) the fitting procedure and which contrast's responses are fit (readings dark, bright, both); (2) "statistically significant positive correlation" as (a) two-sided p < 0.05 with r > 0 or (b) one-sided p < 0.05; (3) "mean and STD" as (a) both below the limit or (b) either. Reproduced when every reading gives a minority of penetrations and sessions (under half of the 59) and one reading puts the counts within tolerance; partially reproduced when the minority holds and no reading matches the counts.
- apc-example-m2p51
In M2p51, 33 neurons well fit by the APC model show mild clustering of angular position peaks (superficial neurons near 330 degrees, deeper ones near 180), with a significant positive correlation between delta AP and distance (r = 0.407) and none in that direction for curvature (r = -0.109).
The distribution of angular position peaks shows mild clustering with superficial neurons (magenta) tuned to features pointing to the lower right (∼330°) and deeper neurons (cyan) tuned to features pointing to the left (∼180°). This relationship can be visualized by the statistically significant positive correlation between inter-neuron distance and difference in angular position tuning peaks (ΔAP) for all pairs of neurons (Fig. 7B). No such relationship is evident along the curvature dimension (Fig. 7C)
Results, Tuning for curvature or angular position, paragraph 1; Figure 7How it will be judged
Method and readings as in apc-peaks-sparse-clustering; M2p51 found as in ex-m2p6-cluster, accepted when its well-fit count is within tolerance. Reproduced when the three values are within tolerance under one reading; partially reproduced when the AP correlation is significantly positive and the CV correlation is not significantly positive, without matching values.
- gyrus-penetrations-consistent
Restricted to the 18 M2 penetrations centered on the V4 gyrus, the percentage of neurons with significant Isimilarity is consistent with the whole population, also when only neurons within about 1,500 um of the most superficial neuron are kept.
When we restricted our analyses to 18 penetrations centered on the V4 gyrus (Fig. 8B, rectangles), our results were consistent withwhat we observedacross the population (Fig. 8C, compare gray and black dots) ... clusters were sparse even when we restricted our analyses to the superficial cortex in those 18 penetrations, e.g., by considering neurons within ∼1,500 μm of the superficial neuron in each penetration
Results, Penetration trajectory, paragraph 1; Figure 8CHow it will be judged
The 18 penetrations are those at the eight grid locations away from both meridians (Figure 8B legend), identified in the paper only graphically (Figure 5A diamonds, Figure 8B); the paper lists neither the sessions nor their grid coordinates, so the claim needs the dataset to identify them. The paper reports no test of "consistent"; it compares the 18 by eye with the population of Figure 5C, whose stated result is the ~20% level of pop-asymptote-20-percent. The 18 are judged against that result. Over those of the 18 with more than 30 neurons, the percentage of neurons with significant Isimilarity, under the readings of pop-asymptote-20-percent, is within its tolerance. A two-sided Mann-Whitney test of per-penetration percentages, the 18 against the other 56, is shown as a variant and does not decide. The 1,500 um restriction recomputes Isimilarity on neurons within 1,500 um of depth zero. Reproduced when the rule holds with and without the restriction.
- texture-sparse-clusters
For textures, of 979 neurons in 51 sessions (6 to 83 per penetration), a small subset (207; 18 in M1, 189 in M2) across 34 of 51 penetrations (7 in M1, 27 in M2) showed clustered selectivity, and 16 penetrations had five or more neurons with significant Isimilarity.
We studied the responses of 979 neurons across 51 sessions (M1, 17; M2, 34) ... (ranging from 6 to 83 per penetration). As with shape stimuli, we found that a small subset of neurons (207/ 979; M1, 18; M2, 189) across 34/51 (M1, 7; M2, 27) penetrations showed clustered selectivity ... Across the 51 penetrations, 16 penetrations had five or more neurons that exhibited regression lines with statistically significant intercepts (Fig. 9A, asterisks)
Results, Population results, texture tuning, paragraph 1; Figure 9AHow it will be judged
As pop-significant-isimilarity, over the 40 textures with original and contrast-reversed versions in place of dark and bright (contrast readings (a) original, (b) contrast-reversed, (c) either), rotations pooled per texture as the Methods present them (rotation is randomized to remove local orientation). The Methods exclude 48 of 979 neurons with too few reliable similarities. Reproduced when one reading puts the counts within tolerance; partially reproduced when the total and penetration counts are and a per-monkey count is not.
- texture-original-reversed-consistent
Texture Isimilarity from original and contrast-reversed textures is consistent across neurons (Spearman's r = 0.464, n = 905).
Tuning similarity based on both sets of texture data (original and contrast-reversed) was highly consistent (Fig. 9B)
Results, Population results, texture tuning, paragraph 1; Figure 9BHow it will be judged
Reproduced when r is within tolerance with p < 0.0001, under either reliability reading.
- texture-asymptote-20-percent
For textures as for shapes, the fraction of neurons with a significant intercept levels off at about 20% when many neurons are recorded.
Our results from texture (Fig. 10, black symbols) are consistent with those from shape (Fig. 10, gray symbols): the fraction of neurons with a significant intercept asymptote to ∼20% when the number of recorded neurons was large
Results, Population results, texture tuning, paragraph 1; Figure 10How it will be judged
As pop-asymptote-20-percent, on the texture penetrations, taking more than 30 neurons as "large", the yield the paper uses for shapes. Readings (a) median of per-penetration percentages and (b) pooled percentage. Reproduced when one reading is within tolerance.
- position-optimal-consistent
In 33 M2 penetrations (723 neurons), Isimilarity computed from responses at each neuron's optimal stimulus position correlates positively with Isimilarity at the aggregate RF center (r = 0.325), so stimulus position and RF scatter do not explain the sparse clusters.
We found a statistically significant positive correlation between Isimilarity in the RF aggregate center versus RF optimal positions (r = 0.325, p < 0.0001, Fig. 12B). These results provide evidence against the possibility that stimulus position and RF variability contributed to the sparsity of clusters
Results, Position invariance for shape tuning similarity, paragraph 1; Figure 12BHow it will be judged
Materials and Methods, RF localization, paragraph 2; per neuron the optimal position is the one of the five with the greatest variance-to-mean ratio of mean rates over the 15 shapes; similarity between two neurons uses each neuron's responses at its own optimal position; Isimilarity as in the opening comment. The Results text says the comparison is with "responses to stimuli at the aggregate RF center" of the position test, while the Figure 12B axis says "Isimilarity (shape test)". The readings are (a) the center position of the position test and (b) the main shape test with dark shapes (the position test's contrast). Spearman's r as Figure 12B names it. Reproduced when every reading gives a positive r with p < 0.0001 and one gives r within tolerance.
- position-examples-invariant
In two example penetrations, the pattern of tuning similarity is the same at all five stimulus positions; a small cluster at every position in M2p9 and none at any position in M2p28.
Results from two representative recording sessions are shown in Figure 11A,C, and in both cases, tuning similarity patterns were similar across all five positions
Results, Position invariance for shape tuning similarity, paragraph 1; Figures 11A and 11CHow it will be judged
The paper shows the matrices and gives no measure. Sessions found as in ex-m2p6-cluster, accepted when the neuron counts match Figure 11 (16 in M2p9, 55 in M2p28) within 10%. Similarity matrices over the 15 shapes at each position, as in the opening comment. OPEN, the paper gives no measure of "similar across positions"; neither reading below is taken from the paper. (a) Pearson's r between the reliable off-diagonal similarities at each displaced position and at the center; holds when all eight correlations (four positions, two penetrations) are positive with p < 0.05. (b) The paper's own cluster measure, Isimilarity (opening comment) at each position; holds when M2p9 has two or more neurons with significant positive Isimilarity at every position and M2p28 has at most one at every position (M1p12, with one such neuron, is the paper's example of no cluster). Reproduced when both readings hold; blocked as methods_underspecified when they differ.
- simulation-random-with-domains
Simulated oblique penetrations through gradual tuning progressions in angular position and curvature give similarity matrices unlike the V4 data, while random tuning peaks interleaved with short stretches of similar tuning reproduce the observed localized clusters.
The resulting similarity matrices (Fig. 13A-C, bottom panels) were quite unlike our V4 data (compare with Figs. 3, 4). Random choice of tuning peaks in the APC space produced similarity matrices reminiscent of the majority of V4 data ... the observed patterns correspond best to random selectivity interspersed with small domains of similar tuning for shapes
Results, Simulation of tuning progression, paragraph 1; Figure 13How it will be judged
A simulation, not a test on recorded data, judged by eye in the paper. Its parameters are partly given (30 model neurons, 2D Gaussian tuning in APC space with SDs drawn from normal distributions of mean 60 degrees, SD 10, and mean 0.5, SD 0.1; Poisson spike counts; three repeats in the Figure 13 legend, 200 ms trains in the Methods); peak amplitudes and the progression step sizes are not. Recorded so the finding is covered; it cannot be tested on the dataset.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none