CRCNS:pfc-1 · Encore Science

Claims 34

From the paper Neural correlates of task switching in prefrontal cortex and primary auditory cortex in a novel stimulus selection task for rodents · claims file · checked 2026-11-06: ready

fig2a-above-chance
Trained rats perform significantly above 50% correct in both localization and pitch discrimination blocks in the recording sessions, with a median of 698 trials per session; the best rats reach about 85% in both blocks. Sessions not above 50% in either block (4 of 55) were discarded.
Trained rats performed many trials per session (median, 698). We verified that the rats were performing significantly above 50% in both blocks ... Our best rats' typical performance during recording sessions was approximately 85% in both blocks ... although some sessions failed the first criterion and were discarded (p > 0.05 for 4/55 sessions)
Results, A Novel Behavioral Task for Rodents, paragraph 2; Figure 2A; Supplemental Experimental Procedures, Chance performance on the task
How it will be judged
Performance per session and block as 2-1-behavres/perf.py computes it, hits over hits plus errors plus wrong-port trials, non-cue blocks 2 and 4, nonrandom 0; sessions as in 2-1-behavres/main00.py (sessions_df.csv include True, less the catch-trial session CR20B_120613_001_behaving). Each session and block is tested against 0.5 with a binomial test (Supplement). Reproduced when at most 4 of the sessions fail in either block, the median trial count is within tolerance and, for at least one rat, the median over its sessions is within tolerance of 85% in both blocks. Open, and named as readings. The trials counted per session, (a) all trials performed (outcome not future_trial), (b) after dropping munged and nonrandom trials. Which 55 sessions the paper's count refers to, (a) the 55 of sessions_df.csv, all marked include, so the four failing sessions may be among them; (b) a set before the cut, so that none of the data set's sessions should fail. Under (b) the claim is reproduced only when no analysed session fails.
figs1a-go-better-than-nogo
Performance is typically better on go trials than on nogo trials.
Performance was typically better on go trials than on nogo trials
Results, A Novel Behavioral Task for Rodents, paragraph 2; Figure S1A
How it will be judged
As 2-1-behavres/main02.py, per rat the mean over sessions of the performance on go and on nogo trials in each block (perf.py, nonrandom 0). Reproduced when go exceeds nogo in at least 7 of the 12 rat and block pairs (a majority, reading "typically"); the pairs are reported.
fig2d-block-switch
After each block change rats switch rapidly to the new target; performance dips briefly during the cue trials at the start of a block and recovers within a few trials.
After each block change, rats rapidly and correctly switched to selecting the new target sound ... Performance briefly dips during cue trials at the beginning of a block but recovers within a few trials
Results, A Novel Behavioral Task for Rodents, paragraph 2; Figure 2D legend
How it will be judged
As 2-1-behavres/main04.py (Figure 2D). Sessions as fig2a-above-chance, nonrandom 0, the first block of each session dropped; trials folded on their position in the 160-trial cycle of a localization and a pitch discrimination block; the fraction of hits at each position, averaged over sessions weighted by trial counts. The paper gives no number for the dip or for "a few trials"; read here, as a choice, as the hit fraction on the first trial of each block lying below the mean over that block's non-cue trials (21 to 80) and reaching that mean within the first 10 trials of the block (within 5 shown as a variant). Reproduced when this holds for both blocks.
iti-median
The median intertrial interval is 4.0 s (interquartile range 2.7 to 5.3 s).
Across the data set, the median intertrial interval was 4.0 s (interquartile range, 2.7-5.3 s)
Results, Within-Trial Timescale of the Encoding of the Selection Rule, paragraph 1
How it will be judged
Neither the paper nor the code defines the interval. Readings, (a) time between consecutive stimulus onsets; (b) time between consecutive center-poke starts; each over all trials of the sessions with included units, pooled. Reproduced when under one reading the median and both quartiles are within tolerance.
fig3c-pfc-rule-encoding
In the hold period of correct trials, 63% (76 of 121) of mPFC neurons individually encode the selection rule; 36 prefer localization and 40 pitch discrimination, and neither preference is significantly more common.
population of mPFC neurons, 63% (76/121) of the neurons individually and significantly encoded the selection rule during the hold period (Figure 3C). Of these, 36 neurons preferred (i.e., fired ... more during) localization and 40 preferred pitch discrimination; neither preference was significantly more common (binomial test, p > 0.05)
Results, Anticipatory Neuronal Activity in mPFC and A1 Encodes the Selection Rule, paragraph 3; Figure 3C
How it will be judged
Units M1, trials M2, test M3 under both readings of the adjustment. Reproduced when under one reading the three counts are within tolerance and a two-sided binomial test of the two preference counts gives p > 0.05.
fig4c-a1-rule-encoding
In the hold period of correct trials, 36% (36 of 99) of A1 neurons encode the selection rule; 13 prefer localization and 23 pitch discrimination, and neither preference is significantly more common.
Across our recorded population, 36% (36/99) of A1 neurons encoded selection rule. As with mPFC, neither population was significantly larger (13 preferring localization, 23 preferring pitch discrimination; binomial test, p > 0.05)
Results, Anticipatory Neuronal Activity in mPFC and A1 Encodes the Selection Rule, paragraph 4; Figure 4C
How it will be judged
As fig3c-pfc-rule-encoding, for the A1 units.
a1-rule-encoding-not-audresp-dependent
In A1, rule encoding is not significantly more common among stimulus-responsive neurons (14 of 49) than among nonresponsive ones (22 of 50).
Rule encoding was not significantly more widespread in either stimulus-responsive (14/49) or nonresponsive (22/50) neurons (p > 0.05, Fisher's exact test)
Results, Anticipatory Neuronal Activity in mPFC and A1 Encodes the Selection Rule, paragraph 4
How it will be judged
A claim of absence. Responsiveness by M4, rule encoding by M3; a two-sided Fisher's exact test on the 2 x 2 table, as 3-1-hold/main3.py. Reproduced when p > 0.05; the counts and the odds ratio are reported.
rule-encoding-effect-size
Among rule-encoding neurons, the median increase in hold-period firing rate in the preferred block is 74.7% in mPFC and 99.7% in A1.
among rule-encoding neurons, the median increase in firing rate during the preferred block was 74.7% in mPFC and 99.7% in A1
Results, Anticipatory Neuronal Activity in mPFC and A1 Encodes the Selection Rule, paragraph 5
How it will be judged
As 3-1-hold/main3.py, per rule-encoding unit the larger of the two ratios of mean hold-period rates (preferred over non-preferred), minus 1; the median over units of each region.
fig3ab-pfc-example-units
Two simultaneously recorded mPFC units encode the rule with opposite preferences; one fires 12.1 Hz in localization against 7.2 Hz in pitch discrimination (483 and 295 trials), the other 5.4 Hz in pitch discrimination against 2.7 Hz in localization, both p < 0.001.
During the hold period (gray), the firing rate is significantly (p < 0.001, unpaired Mann-Whitney U-test) higher on localization (mean 12.1 Hz, n = 483) than on pitch discrimination (mean 7.2 Hz, n = 295) trials ... The hold period firing rate is significantly (p < 0.001) higher on pitch discrimination (mean 5.4 Hz) than on localization (mean 2.7 Hz) trials
Figure 3A and 3B legends
How it will be judged
The units are named in metadata/figures_generated_by_each_script.csv, 3A CR24A_121021_001_behaving-257 and 3B CR24A_121021_001_behaving-259. Hold rates and test as M2 and M3 (250 ms window); reproduced when the rates and trial counts are within tolerance and each unadjusted test gives p < 0.001.
fig4ab-a1-example-units
Two simultaneously recorded A1 units encode the rule; one fires 8.0 Hz in localization against 4.8 Hz in pitch discrimination, the other 10.1 Hz in pitch discrimination against 2.0 Hz in localization, both p < 0.001.
An example A1 neuron that responds significantly more (p < 0.001) during localization (8.0 Hz, n = 312 trials) than during pitch discrimination (4.8 Hz, n = 253) ... Simultaneously recorded A1 neuron that significantly (p < 0.001) prefers pitch discrimination (10.1 Hz, n = 312 trials) over localization (2.0 Hz, n = 253)
Figure 4A and 4B legends
How it will be judged
Units as named in metadata/figures_generated_by_each_script.csv, 4A CR20B_120602_001_behaving-612 and 4B CR20B_120602_001_behaving-712; method as fig3ab-pfc-example-units. The two legends give the same session's trial counts with the blocks swapped (312 localization and 253 pitch discrimination in 4A, the reverse in 4B), so the counts are compared as a pair, 312 and 253 within 10% each, in either assignment, and the assignment found is reported.
fig3d-pfc-interference
In mPFC rule-encoding neurons, rule encoding is weakened on interference trials; the hold-period response on interference trials is higher than on correct trials in the non-preferred block and lower in the preferred block (57 neurons).
in mPFC, the encoding of selection rule was significantly weakened on interference trials, as compared with correct trials ... The population response on interference trials is significantly increased during the nonpreferred block and decreased during the preferred block (n = 57 neurons, paired Mann-Whitney test)
Results, Error Trial Analysis, paragraph 2; Figure 3D legend
How it will be judged
As 3-2-errhold/main1.py and main2.py. Hold counts as M3 on all trials of blocks 2 and 4 with nonrandom 0, by stimulus and outcome. Categories per block, hits; interference trials, wrong-port responses to stimuli whose distractor cues a go (localization le_lo_lc and ri_lo_lc, pitch discrimination le_hi_pc and le_lo_pc); go-on-nogo errors (localization ri_hi_lc and ri_lo_lc, pitch discrimination le_hi_pc and ri_hi_pc). Rule-encoding units (M3) with at least 3 trials in each of the six categories; rates in Hz relabelled by preferred block and normalized by subtracting the hit rate in the non-preferred block. Wilcoxon signed-rank tests (M5) across neurons, hit against interference in the non-preferred block and in the preferred block. Reproduced when both are significant at p < 0.05 with interference higher in the non-preferred and lower in the preferred block; partially reproduced when one holds.
fig4d-a1-interference
In A1 rule-encoding neurons the rule encoding is reversed on interference trials; the hold-period response on interference trials is greater in the non-preferred block than in the preferred block (16 neurons).
The population response on interference trials (orange bars) is significantly greater during the nonpreferred block than during the preferred block (p < 0.01, n = 16 neurons, paired Mann-Whitney U-test)
Results, Error Trial Analysis, paragraph 2; Figure 4D legend
How it will be judged
As fig3d-pfc-interference, for A1 units; the test is interference in the preferred block against interference in the non-preferred block (3-2-errhold/main2.py). Reproduced when the non-preferred interference response is higher with p < 0.01.
figs2c-s3c-hold-by-trial-type
In rule-encoding neurons of both regions, hold-period activity does not differ between correct go, correct nogo and go-on-nogo error trials within a block, while interference trials differ from all other trial types.
There is no significant difference in the hold period activity between correct go, correct nogo, and incorrect go-on-nogo trials ... However, there is a significant difference between interference trials and all other trial types in that block ... However, there is no difference between the other trial types, regardless of whether the rat performed a go or nogo response
Figure S2C and S3C legends
How it will be judged
In part a claim of absence. As 3-2-errhold/main3.py and the Figure S2C legend, which names the test (paired Mann-Whitney U-test across neurons, Bonferroni corrected) but prints no p-value, so no statistics entry is recorded. Rule-encoding units (M3), hold counts as M3 on trials of blocks 2 and 4 with nonrandom 0, in four categories per block, correct go, correct nogo, go-on-nogo errors and interference trials (stimuli as fig3d-pfc-interference); units with at least 3 trials in each of the eight categories; rates relabelled by preferred block. Wilcoxon signed-rank tests (M5) between every pair of categories within each block, Bonferroni corrected over all comparisons of a region. Reproduced when, in each region, no comparison among the three non-interference categories has corrected p < 0.05 and every comparison of interference with another category in the same block has corrected p < 0.05; partially reproduced when only the absence holds. The p-values are reported.
figs2f-s3f-bidirectional-vs-spont
Across rule-encoding neurons in both regions, the hold-period rate is significantly above the spontaneous rate in the preferred block and significantly below it in the non-preferred block.
across the population of rule-encoding neurons, the firing rate was significantly higher than spontaneous during the preferred block and significantly lower than spontaneous during the nonpreferred block
Results, Within-Trial Timescale of the Encoding of the Selection Rule, paragraph 4; Figures S2F and S3F
How it will be judged
As 3-3-vs_spont/main2.py and main3.py. Spontaneous rate of each unit, the spike count over the time between its session's first and last stimulus onset that lies more than 2 s from every stimulus onset (all trials, munged included, time in 0.1 s steps); hold rates as M3. Wilcoxon signed-rank tests (M5) of hold rate against spontaneous rate, in each region for the localization-preferring and the pitch-discrimination-preferring units separately, each block. Reproduced when, in all four region and preference groups, the preferred-block rate is above and the non-preferred-block rate below the spontaneous rate at p < 0.05; partially reproduced when this holds in some groups. The paper's text pools preferences, so the pooled test per region (preferred and non-preferred relabelled) is shown as a variant.
fig5ab-single-unit-onset
The median rule-encoding unit first shows a significant block preference 0.55 s before the stimulus in A1 and 0.625 s before it in mPFC, so most develop it before the trial is initiated.
In A1, the median rule-encoding unit first developed a significant block preference 0.55 s prestimulus; in PFC, the median was 0.625 s prestimulus. Thus, the majority of ... neurons developed this property well before the rat initiated a trial by center-poking
Results, Within-Trial Timescale of the Encoding of the Selection Rule, paragraphs 1 and 2; Figures 5A and 5B
How it will be judged
Per rule-encoding unit, M2 trials, spike trains smoothed with a 50 ms Gaussian and sampled in 50 ms bins (Results; Figure 5A legend; 5-2-timecourse_exemplars/main1.py), an unpaired Mann-Whitney test between blocks in each bin, p-values adjusted by Benjamini-Hochberg over all bins and units together (main2.py, main3.py); the interval is the contiguous run of significant bins, with the preferred direction, that contains the bin just before stimulus onset, and its first bin is the onset; median over units per region (main4.py). The tolerance is one 50 ms bin, the resolution of the measure. The window differs, the Results say up to 3 s before and after onset, the code uses 5 s; the Results' 3 s is the main reading and 5 s a variant. Reproduced when both medians are within tolerance and both exceed 0.35 s, the longest hold.
fig5c-population-onset
The population activity of rule-encoding neurons in their preferred block ramps up before the stimulus and is significantly elevated earlier in mPFC (from 2.7 s before stimulus onset) than in A1 (from 0.88 s).
The PFC population increases its response earlier (first significantly activated 2.7 s before stimulus onset) than the A1 population (first significantly activated 0.88 s before stimulus onset)
Results, Within-Trial Timescale of the Encoding of the Selection Rule, paragraph 3; Figure 5C legend
How it will be judged
As 5-1-timecourse/main1.py and main2b.py and the Supplement (Timecourse analysis). Hits on the eight non-cue stimulus pairs, nonrandom 0, locked to stimulus onset; 50 ms bins from -2.9 to 2.9 s, 50 ms Gaussian smoothing; each unit's rates z-scored over all bins and stimulus pairs, averaged within go and nogo pairs of each block and then over go and nogo; the preferred block of each rule-encoding unit. Per bin a one-sample t-test against 0 over units, Benjamini-Hochberg over bins; the first significant bin with positive mean is the onset (bin centre). Reproduced when both times are within tolerance and the mPFC time is earlier; partially reproduced when only the order holds.
fig5d-go-nogo
In the preferred block, after the rat leaves the center port, the population rate of rule-encoding neurons stays above baseline for several seconds on nogo trials and falls below baseline on go trials, in both mPFC and A1.
On nogo trials (gray), the firing rate remained elevated above baseline for at least several seconds ... On go trials, the firing rate was suppressed below baseline as the rat moved to the choice port and consumed a reward
Results, Encoding of Behavioral Choice, paragraph 1; Figure 5D legend
How it will be judged
As 5-1-timecourse/main1.py and main2.py, locked to center-port exit (cpoke_stop), 50 ms bins from -3 to 3 s, counts without smoothing, z-scored per unit as in fig5c-population-onset (baseline is 0); the preferred block; go and nogo pairs kept apart. The paper reports no test or window for this panel. Readings, (a) the mean over units of the mean z-scored rate from 0 to 3 s after exit, tested against 0 with a one-sample t-test over units (the test of Figure 5C); (b) per-bin one-sample t-tests with Benjamini-Hochberg over bins, as for Figure 5C, the claim holding when nogo is significantly above 0 and go significantly below 0 over most bins of 0.5 to 3 s. Reproduced when under one reading nogo is above and go below baseline at p < 0.05 in both regions; partially reproduced when this holds in one region.
audresp-prevalence
Auditory-responsive neurons are significantly rarer in mPFC (31 of 121) than in A1 (49 of 99).
Auditory-responsive neurons in mPFC were significantly rarer (mPFC, 31/121; A1, 49/99; p < 0.001, Fisher's exact test) than in A1
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 1
How it will be judged
Responsiveness by M4 (both readings), two-sided Fisher's exact test as 6-2-evokmisc/main1.py. Reproduced when under one reading both counts are within tolerance and p < 0.001.
figs5cd-evoked-strength
Onset responses are stronger in auditory-responsive A1 neurons than in mPFC neurons, 0.12 against 0.02 extra spikes per trial (median) and 209% against 171% of baseline.
The responses are significantly stronger in A1 neurons (median: 0.12 spikes) than in PFC neurons (median: 0.02 spikes), p < 0.001, unpaired Mann-Whitney U-test ... Again the responses are stronger in A1 neurons (median: 209% of baseline) than in PFC neurons (median: 171% of baseline), p < 0.001, unpaired Mann-Whitney U-test
Figure S5C and S5D legends
How it will be judged
As 6-2-evokmisc/main0.py and main1.py. All trials and stimuli of each responsive unit (M4); baseline rate from the 50 ms before onset, evoked rate in the response window; extra spikes are (evoked rate minus baseline rate) times the window length, percent is evoked over baseline rate. Reproduced when the four medians are within tolerance and both tests give p < 0.001.
figs5e-latency
The median response latency of auditory-responsive mPFC neurons is slightly but significantly longer than in A1, 19.75 against 16.75 ms.
Their median response latency was also significantly longer, though only slightly (19.75 ms versus 16.75 ms in A1; Figure S5E) ... though only slightly. p < 0.05, unpaired Mann-Whitney U-test
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 1; Figure S5E legend
How it will be judged
Latency is the centre of the response window (Figure S5E legend; 6-2-evokmisc/main1.py takes the mean of audresp_t1 and audresp_t2). Reproduced when both medians are within tolerance, the mPFC median is the longer and the two-sided test gives p < 0.05.
fig6cd-baseline-evoked-correlation
Across auditory-responsive neurons, the change in hold-period rate between blocks correlates with the change in evoked rate, in A1 (43 neurons, trend slope close to 1) and in mPFC (17 neurons), p < 0.001 in each.
For A1 neurons (n = 43), increased anticipatory firing during one block significantly correlated (p < 0.001) with increased evoked responses during the same block ... The slope of the trend line is close to 1 ... but for auditory-responsive mPFC neurons (n = 17). Again, the change in anticipatory activity across neurons correlated closely with the change in evoked response (p < 0.001)
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 2; Figure 6C and 6D legends
How it will be judged
As 6-1-evoked/main0.py and 6-3-evokgain/main0.py, main2.py. Evoked counts in the response window (M4) on hits of the eight non-cue stimulus pairs, nonrandom 0. Units are dropped when a one-sample t-test of all their evoked rates against their hold rate (block-weighted) gives p > 0.05. Evoked change is the stimulus-equalized difference, pitch discrimination minus localization, of mean evoked rates (mean within each stimulus pair, then over pairs); baseline change is the hold-rate difference of M3. Linear regression of evoked change on baseline change (scipy linregress). Reproduced when both regressions are positive with p < 0.001 and the unit counts are within tolerance. "Close to 1" is given no value; the A1 slope is reported with its 95% confidence interval, and taken to hold when that interval contains 1, a choice made here.
evoked-beyond-baseline-rare
After subtracting each block's baseline, only 6 of 43 A1 neurons (14%) change their evoked response significantly between blocks, 4 of the 6 from one rat (Rat 1), with higher rates in pitch discrimination.
we found very little evidence of any block-specific change in evoked firing rate ... We found that a small population (6/43, or 14%) of A1 neurons increased their evoked response significantly ... these neurons were predominantly (4/6) observed in a single animal (Rat 1)
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 2; Supplemental Experimental Procedures, Changes in magnitude of evoked response not explained by changes in baseline
How it will be judged
As 6-3-evokgain/main0.py with the block's hold rate subtracted from each trial's evoked rate; trials drawn with replacement equally from each stimulus pair in each block (10000 draws, at least 5 trials per stimulus), two-sided bootstrap p of a zero difference, floored at 1/10000, Benjamini-Hochberg within region (the code; the Supplement gives p < 0.05 and names no correction, so the unadjusted count is shown as a variant). Reproduced when the count is 5 to 7 of the A1 units, 3 to 5 of them come from CR12B and these are a majority of the count, and the CR12B units have the higher baseline-subtracted evoked rate in pitch discrimination; partially reproduced when the count is within 3 to 9. The Supplement's "in these neurons the firing rate was higher during pitch discrimination" is read as the CR12B units; the direction of all significant units is reported. The tolerance of one neuron on 6 and 4 is the 10% rule rounded up to a whole neuron.
fig6e-decoding
Sound identity is decoded more accurately from A1 ensembles (21 ensembles, 49 neurons) than from mPFC ensembles (17 ensembles, 31 neurons), but for neither sound nor region is decoding better in one block than the other.
It could be decoded significantly more accurately (p < 0.001 for the main effect of brain region) using A1 responses (n = 21 ensembles of 49 neurons total) than using mPFC responses (n = 17 ensembles of 31 neurons total). However, for both sounds and both brain regions, the decoding was not significantly more accurate during either pitch discrimination (red) or localization (blue) trials (p > 0.05 for each pair of bars using a paired t test)
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 3; Figure 6E legend
How it will be judged
As 6-1-evoked/main1.py, main2.py, main4.py and the Supplement (Decoder analysis). An ensemble is the responsive units (M4) of one region in one session. Per ensemble, block and sound (noise burst side, warble pitch), a logistic regression (scikit-learn, class weights balanced) on the trial-by-trial evoked counts of hits of the four stimulus pairs, scored in sample as the fraction correct averaged over the two classes. A type II three-way ANOVA of the scores on region, sound and block; paired t-tests between blocks per region and sound, adjusted by Benjamini-Hochberg as the code does (the paper does not say). Reproduced when the region effect has p < 0.001 with A1 higher and all four block comparisons give p > 0.05; the region claim and the block claim (an absence, judged on the paired t-tests) are reported apart. Leave-one-out scores are shown as a variant (Supplement).
figs5fg-tuning-vs-rule
There is no correlation between a neuron's change in anticipatory activity and its tuning for the noise burst or the warble, in A1 or mPFC.
we did not observe any correlation between each neuron's change in anticipatory activity and its tuning for the stimuli
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 4; Figure S5F and S5G legends
How it will be judged
A claim of absence. As 6-1-evoked/main2.py and main6.py and the Figure S5F legend. Tuning is the single-unit decoding score (as fig6e-decoding) of the cue sound on cue-trial hits, the noise burst from localization cue trials and the warble from pitch discrimination cue trials; units with at least 25 cue trials in each block and at least 10 evoked spikes on them; anticipatory change is log10 of the pitch discrimination over localization hold rate. Linear regression of the change on each tuning, per region. Reproduced when all four regressions give p > 0.05; the r values are reported.
figs5hi-stimulus-pairs
No stimulus pair shows a greater change in evoked response between blocks than the others, with or without baseline subtraction, and in absolute value; and no stimulus pair's change differs from zero.
A one-way Kruskal-Wallis test revealed no significant difference between the stimulus pairs ... We found no significant difference from zero for any individual stimulus pair
Results, Changes in Baseline Activity Correlate with Changes in Evoked Activity, paragraph 4; Figure S5H and S5I legends
How it will be judged
A claim of absence, as 6-1-evoked/main5.py. Per responsive unit (M4) and stimulus pair, the square root of the mean evoked rate on hits in pitch discrimination minus that in localization; baseline-subtracted as the square root of the evoked rate minus the square root of the block's hold rate. Kruskal-Wallis across the four pairs per region, raw and baseline-subtracted, and on absolute values; Wilcoxon signed-rank against zero per pair with Benjamini-Hochberg over the four pairs. Reproduced when no test gives p < 0.05; the p-values are reported.
figs2gh-s3gh-rt-mt
Anticipatory firing rate does not correlate with reaction time or movement time; few neurons correlate individually and the mean correlation does not differ from zero.
we did not observe any correlation between the anticipatory firing rate and the reaction or movement time
Discussion, The Potential Roles of Motor Planning and Posture, paragraph 3; Figures S2G, S2H, S3G, S3H legends
How it will be judged
A claim of absence, as 3-4-RT_vs_hold/main.py and main3.py. Per unit and block, the correlation (linear regression) between the square root of the hold count and reaction time (center-port exit minus stimulus onset) on all hits, and movement time on go hits. The code takes movement time as choice-port entry minus stimulus onset; the legend's "time necessary for the rat to move to the choice port" can also be read as choice-port entry minus center-port exit, shown as a variant. Reproduced when the one-sample t-test of the r values against zero gives p > 0.05 in each region, block and measure; the count of individually significant units (Benjamini-Hochberg) is reported.
figs2i-head-angle
The rat's head angle during the center poke differs between blocks by about 30 degrees, 26.9, 36.6 and 24.1 degrees in the three scored sessions.
differs by approximately 30 degrees between blocks ... The mean difference between blocks was 26.9, 36.6, and 24.1 degrees in each of the three analyzed sessions
Results, Potential Role of Posture, paragraph 1; Figure S2I legend; Supplemental Experimental Procedures, Video analysis of preparatory head positioning
How it will be judged
From the ear positions in miscellaneous/vidtrack, head angle as vidtrack.vts_db2head_pos_df computes it (in degrees), on the trials of the hold analysis (M2), as 3-5-vidtrack/main0.py and main1.py. The paper does not say which value belongs to which session, so the three differences of block means are compared to the three reported values as a set (sorted). Reproduced when all three are within tolerance.
figs2kl-s3kl-fev
For most rule-encoding neurons, block explains far more of the hold-period variance than head angle; the median fraction explained by block is 94.1% in mPFC and 83.0% in A1 (head angle 5.9% and 17.0%), and only 3 of 16 mPFC and 1 A1 neuron are mostly explained by head angle.
Only three out of 16 bars are more than 50% black ... most of the variance (median: 94.1%) is explained by block; the small amount that remains is explained by head angle (median: 5.9%) ... Only one bar is more than 50% black ... most of the variance (median: 83.0%) is explained by block; the small amount that remains is explained by head angle (median: 17.0%)
Figures S2K, S2L, S3K, S3L legends; Supplemental Experimental Procedures, Video analysis, Results
How it will be judged
Rule-encoding units of the three video sessions (16 mPFC, 8 A1); per unit a type II ANOVA of the square root of hold counts on head angle plus block (3-5-vidtrack/main7.py); FEV of a factor is its sum of squares over the sum for both. One A1 unit whose overall F-test gives p > 0.05 is left out (Supplement, technical note). Reproduced when the medians are within tolerance and the counts of units with head-angle FEV above 50% are 3 (mPFC) and 1 (A1). Tolerances on the FEV medians are 10% of each value, so block and head angle are tested apart.
posture-equalized-sign
On posture-equalized trials the rule encoding persists; its population spread (0.471 in mPFC, 0.318 in A1) exceeds that of shuffles (p < 0.005), and only 1 of 16 mPFC and 0 of 8 A1 cells switch preferred block (binomial p = 0.0005 and 0.008).
The PRE was 0.471 in PFC and 0.318 in A1 on the posture-equalized trials; the fraction of shuffles that yielded a greater PRE was <0.005 in both brain regions ... In fact we observed that only 1/16 PFC cells and 0/8 A1 cells switch their preferred block after posture equalization. The p-value (two-tailed binomial test) of this result is 0.008 for A1 and 0.0005 in PFC
Results, Potential Role of Posture, paragraph 2; Figures S2M, S2O, S3M, S3O; Supplemental Experimental Procedures, Posture equalization by trial selection
How it will be judged
As 3-5-vidtrack/main3.py to main5.py. Posture-equalized trials per session, the localization trials with the highest head angles and the pitch discrimination trials with the lowest, the largest equal number for which the localization mean stays above the pitch discrimination mean, less one. Rule encoding is log10 of the pitch discrimination over localization mean count; PRE is the scaled median absolute deviation (statsmodels stand_mad) over units; 10000 shuffles of block labels on the posture-equalized trials. Reproduced when the PREs are within tolerance with shuffle p < 0.005 and the switch counts match. Switch counts are whole cells and taken exactly.
posture-equalized-magnitude
The mean magnitude of rule encoding is not significantly altered on posture-equalized trials, in mPFC or A1.
we found that the mean magnitude of rule-encoding was not significantly altered by postural equalization
Figures S2N and S3N legends; Supplemental Experimental Procedures, Posture equalization by trial selection
How it will be judged
A claim of absence. Absolute rule encoding (as posture-equalized-sign) on posture-equalized and on the other trials, compared per region with the Wilcoxon signed-rank test (M5). Reproduced when p > 0.05 in each region; the change is reported.
fig7-disruption-by-block
Electrical disruption of mPFC impairs performance; across sessions the impairment is significant in pitch discrimination for all three rats and in localization for two of three.
This impairment was significant across sessions (p < 0.05, binomial test on the number of impaired sessions) during pitch discrimination for all (three of three) rats and during localization for most (two of three) rats
Results, Disruption of mPFC Significantly Impairs Task Performance, paragraph 2; Figure 7A
How it will be judged
As 7-1-zap/zap.py and main1.py on miscellaneous/zap/TI_d. Per session, performance (fraction of hits) on disruption and control trials of block 2 and of block 4, munged trials dropped (the code does not drop nonrandom trials; dropping them is a variant). Per rat, a two-sided binomial test of the number of sessions with control above disruption performance. Reproduced when pitch discrimination is significant for all three rats and localization for two; partially reproduced when at least four of the six rat and block tests are. Session counts are whole sessions and taken exactly.
fig7c-disruption-nogo
Disruption impairs nogo trials in particular; every rat is significantly impaired on nogo trials in one block or the other, and Z2 also on localization go trials.
All rats were significantly impaired on nogo trials during one block or the other. One rat (Z2) also showed a significant impairment on localization go trials
Results, Disruption of mPFC Significantly Impairs Task Performance, paragraph 2; Figure 7C legend
How it will be judged
As fig7-disruption-by-block, split by go and nogo in each block (zap.py compute_by_block_and_gonogo). Reproduced when each rat has a significant nogo impairment in at least one block and, among go trial types, only Z2 localization go is significant.
figs1bc-catch
Rats perform as well when the distractor is replaced by a novel neutral sound as on standard trials, so they select the target and do not memorize the four stimulus pairs.
The rats perform just as well on catch trials as on standard trials (Fisher's Exact Test on the trial outcomes, p > 0.05 in all cases)
Figure S1B and S1C legends
How it will be judged
A claim of absence, as 2-1-behavres/main05.py on miscellaneous/catch_trial_task. For each rat and each catch stimulus, hits and trials on it against the two standard stimuli with the same target pooled; two-sided Fisher's exact test. Reproduced when all eight tests give p > 0.05; the performances are reported.
fig8-model
In the network model with 320 neurons, a task signal alone lets the network perform above 80% correct with a sensory signal-to-noise ratio as low as 0.0625, and 100% at an SNR of 0.25 for virtually any positive task signal.
With 320 neurons, the network performs above 80% correct even with a signal-to-noise ratio (SNR) as low as 0.0625 ... At the highest SNR of 0.25 (darkest trace), the model produces 100% correct responses for virtually any positive task switch signal
Results, A Simulated Network Model, paragraph 3; Figure 8B legend
How it will be judged
From the deposited simulation results (miscellaneous/model, simulation_resdf and simulation_params.pickle), as 8-1-model/main1.py; performance is score1b or score2b over NT_TEST for the task whose population receives the signal. SNR is the sensory gain over the noise level; the Supplement calls the noise its strength and Figure S7 its variance, so both gain/noise and gain/sqrt(noise) are named, the one that yields the printed SNR grid (powers of 2 from 2^-6 to 2^-2) taken. Reproduced when for N = 320 and SNR 0.0625 some positive task signal gives above 80% correct in both tasks, and at SNR 0.25 performance is at least 99% over most positive task signals. The SNR is a grid value and taken exactly.

Pipeline

Paper: waiting (not checked)
Claims: done
Claims check: done (ready)
Attempt: todo
Questions: none
Verified: none