Claims 36
From the paper Preprint manuscript · claims file · checked 2026-10-23: ready
- dataset-neuron-counts
The passive identity dataset holds 81 OB-p, 84 mPFC-p and 1978 untagged neurons from 10 mice; the concentration dataset 87 OB-p, 102 mPFC-p and 2232 untagged neurons from 10 mice; the Go/NoGo dataset 7 mice (4 OB-p, 3 mPFC-p tagged) with, per session, 95 +/- 2 OB-p, 87 +/- 8 mPFC-p and 1690 +/- 35 untagged neurons.
head-fixed mice were either exposed to 8 monomolecular odorants at a single concentration (10 mice, totaling 81 OB-p, 84 mPFC-p, and 1978 untagged aPCx neurons ... or to three odorants at three concentrations each (10 mice, totaling 87 OB-p, 102 mPFC-p, and 2232 untagged cells ... Per session (mean ± SD), we recorded 95 ± 2 OB-p, 87 ± 8 mPFC-p, and 1690 ± 35 untagged neurons
Results, Calcium imaging of piriform neurons with distinct projection targets, paragraph 2How it will be judged
Counts of suite2p cells (iscell) by tag, per experiment, as the code's load_experiment builds them. The Go/NoGo tolerances are the paper's SD. "Per session" is read as (a) the total over all mice of one task day, averaged over task days 1 to 3, or (b) the same over days 0 to 3; for OB-p and mPFC-p, only the mice tagged for that target count. Reproduced when all values are within tolerance under one reading; partially reproduced when the identity and concentration counts are and the Go/NoGo counts are not.
- fast-dataset-counts
The 30 Hz subset holds 24 OB-p, 27 mPFC-p and 424 untagged neurons from 5 mice.
in a subset of experiments (5 mice, totaling 24 OB-p, 27 mPFC-p, and 424 untagged cells), we examined the temporal structure of odor responses by performing high-speed (30 Hz) imaging
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 3 - figs2b-labeling-similar
OB-p and mPFC-p mice do not differ in the number of jGCaMP7f neurons or of co- labeled jGCaMP7f/tdTomato neurons.
Quantification of jGCaMP7f and co-labeled jGCaMP7f/tdTomato-expressing neurons in OB-p and mPFC-p mice. (OB-p, N = 9; mPFC-p, N = 10. Unpaired two-tailed t-test
Figure S2B legendHow it will be judged
A claim of absence, judged on the paper's test. Per mouse, the number of suite2p cells and of red (tdTomato-tagged) cells, with an unpaired two-tailed Student t-test of OB-p against mPFC-p mice for each count. The legend does not say which session gives a mouse's count when a mouse was imaged more than once; readings are (a) each mouse's first imaging session and (b) its mean over sessions. The legend also leaves open whether the co-labeled measure is a count or a share of the mouse's jGCaMP7f neurons; readings are (c) the count and (d) the percentage of jGCaMP7f neurons. Reproduced when neither test has p < 0.05 under the readings; not reproduced when either has p < 0.05 under all readings. The mouse counts per group are reported against 9 and 10.
- mlp-projection-f1
An MLP trained on normalized fluorescence traces pooled across the identity and Go/NoGo datasets classifies OB-p against mPFC-p neurons with F1 scores of 0.688 (OB-p) and 0.557 (mPFC-p).
The MLP achieved F1 scores of 0.688 (OB-p) and 0.557 (mPFC-p)
Results, Temporal structure of activity predicts OB-p vs mPFC-p identity, paragraph 1How it will be judged
Inputs as for the transformer (Methods, Preprocessing), trials from 1 s before to 15 s after odor onset (70 frames), L = 5 subsamples of M = 20 trials per neuron, the identity passive and Go/NoGo datasets, within-animal split 0.8/0.1/0.1 of neurons, averaged over the seeds 0, 1, 2, 3, 4, 42, 43, 44, 45, 46. The MLP's architecture and training hyperparameters are in Table 8, which is missing from the text; until it is supplied the claim is blocked as methods_underspecified. Whether untagged neurons are a third class in training (Figure 1F colors them) is open; readings are (a) two classes, OB-p and mPFC-p only, and (b) three classes, with F1 reported for OB-p and mPFC-p.
- transformer-projection-f1
A transformer classifier on single-neuron calcium traces decodes projection target with F1 0.759 (OB-p) and 0.744 (mPFC-p) when neurons from all animals are split, and 0.754 (OB-p) and 0.623 (mPFC-p) when one OB-p and one mPFC-p animal are held out.
which improved test performance to 0.759 (OB-p) and 0.744 (mPFC-p) ( Figure S3B ) ... Performance remained high ... 0.754 OB-p; 0.623 mPFC-p
Results, Temporal structure of activity predicts OB-p vs mPFC-p identity, paragraph 1; Figure S3BHow it will be judged
As in the Methods (Projection decoding analysis with a transformer model). Identity passive and Go/NoGo datasets; trials from 1 s before to 15 s after odor onset (70 frames); L = 5 subsamples of M = 20 trials per neuron; patches of 2 frames embedded in 128 dimensions; a ViT encoder with a class token; trial latents average-pooled; majority vote over subsamples; Adam with cosine decay. Within-animal split 0.8/0.1/0.1 over seeds 0, 1, 2, 3, 4, 42, 43, 44, 45, 46; across-animal over 6 held-out OB-p/mPFC-p pairs (Table 6) with seeds 42, 43, 44. The depth, heads, learning rate, epochs and batch size (Table 5) and the held-out pairs (Table 6) are missing from the text; until they are supplied the claim is blocked as methods_underspecified. Classes as in mlp-projection-f1 (readings a and b).
- fig1g-attention-timing
Attention rollout on correctly classified neurons puts the informative epochs early after odor onset (about 0 to 3 s) for OB-p neurons and later in the trial for mPFC-p neurons.
informative epochs cluster early after odor onset (around 0-3 s) for OB-p neurons, but later in the trial for mPFC-p neurons
Results, Temporal structure of activity predicts OB-p vs mPFC-p identity, paragraph 2; Figure 1G and Figure S3DHow it will be judged
From the transformer of transformer-projection-f1, attention rollout (Abnar and Zuidema 2020) of the class token over patches, averaged over trials per correctly classified subsample, averaged over 10 seeds, OB-p minus mPFC-p (Figure 1G legend), on the identity dataset. Reproduced when the difference is positive over most of 0 to 3 s after odor onset and negative over most of the time after 3 s. Blocked with the transformer while Table 5 is missing.
- fig1f-latent-separation
The transformer's penultimate-layer embeddings separate OB-p from mPFC-p neurons, with little clustering by animal or dataset.
which separated OB-p and mPFC-p well with minimal clustering by animal or dataset
Results, Temporal structure of activity predicts OB-p vs mPFC-p identity, paragraph 2; Figure 1F and Figure S3E to S3GHow it will be judged
The paper shows PCA plots and gives no measure of separation, so the claim can be judged only by inspecting the same plots (PCA of neuron embeddings, seed 42, colored by label, animal and dataset). It is blocked with the transformer while Table 5 is missing; if run, it is reported as plots without an outcome beyond what they show.
- fig2b-lifetime-sparseness-similar
Lifetime sparseness does not differ significantly between OB-p, mPFC-p and untagged neurons in the identity dataset.
Odor responses in PCx were generally sparse on the single cell level, with no significant differences in lifetime sparseness between groups
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 2; Figure 2BHow it will be judged
A claim of absence. Per neuron, lifetime sparseness over the 8 odors of its trial- averaged binarized activity (0 to 5 s window), 81 OB-p, 84 mPFC-p and 1978 untagged neurons. The text names no test here; the test the paper uses for the other single-cell comparisons of Figure 2 is the two-sided Mann-Whitney U test with Bonferroni correction, taken here over the 3 pairs. Reproduced when no pair has corrected p < 0.05; not reproduced when any does. Medians are reported.
- fig2c-mutual-information
Mutual information between binarized activity and odor identity does not differ between OB-p and mPFC-p neurons, but is lower for OB-p than untagged neurons (p = 0.012).
Mutual information did not differ significantly between projection sub-populations, but was lower for OB-p neurons than for untagged neurons (p = 0.012)
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 2; Figure 2CHow it will be judged
Per neuron, mutual information between its binary activity per trial (0 to 5 s window) and the odor (8 odors). Two-sided Mann-Whitney U tests with Bonferroni correction over the 3 pairs. Reproduced when OB-p against mPFC-p has corrected p >= 0.05 and OB-p is lower than untagged with corrected p < 0.05.
- fig2d-population-sparseness
Population sparseness per trial differs only between OB-p and untagged populations (p = 1.26 x 10^-5).
At the ensemble level, population sparseness differed only between OB-p ensembles and the untagged populations
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 2; Figure 2DHow it will be judged
Population sparseness per pseudo-population trial (8 odors by 8 trials, 64 per group) over all cells of the group, from their binary activity on that trial (0 to 5 s). Two-sided Mann-Whitney U tests over the 3 pairs, Bonferroni corrected. Reproduced when only OB-p against untagged has corrected p < 0.05. The direction is reported (the paper does not state it).
- fig2-proportion-responding
A higher proportion of OB-p than mPFC-p neurons responds to odorants.
with OB-p neurons responding in a significantly higher proportion than mPFC-p neurons (two-sided Mann–Whitney U test with Bonferroni corrected p-values, p = 5.79 × 10⁻ 7 )
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 2How it will be judged
The proportion of active cells (Methods, Metrics derived from response binarization) per trial, 0 to 5 s window. Readings of the sample are (a) per pseudo- population trial over all cells of the group (64 values per group) and (b) per neuron, the proportion of trials it is active (mean binarized activity). Two-sided Mann-Whitney U test, Bonferroni corrected over the 3 pairs. Reproduced when OB-p exceeds mPFC-p with corrected p < 0.05 under one reading.
- fig2e-reliability
OB-p populations contain higher fractions of both unreliable and reliable responsive cells per trial (10.97% and 1.54%) than mPFC-p populations (8.71% and 0.25%).
OB-p populations contained higher fractions of both unreliable and reliable cells (10.97% and 1.54%, respectively) than mPFC-p populations ( Figure 2E , 8.71% and 0.25%
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 2; Figure 2EHow it will be judged
Methods, Reliability analysis. A cell responds on a trial when its mean normalized fluorescence 0 to 5 s after onset exceeds its baseline mean plus 2.33 baseline SDs; a cell is reliable if it responds on more than half of its odor trials, unreliable otherwise; per trial, the percentage of the group's cells that respond and are reliable, and that respond and are unreliable. Open details are the baseline window, read as (a) the 4 s before onset (the Methods' baseline for active responses) or (b) the whole 5 s pre-odor baseline; whether the baseline mean and SD are per trial or pooled over the cell's trials, read as (c) per trial or (d) pooled; and whether "more than half" counts over all odor trials or per odor, read as (e) all odor trials or (f) any odor. Trials are pseudo- population trials (64). Two-sided Mann-Whitney U tests of OB-p against mPFC-p, Bonferroni corrected. Reproduced when, under one combination of readings, the four percentages are within tolerance and both tests have corrected p < 0.05 with OB-p higher; partially reproduced when both directions and tests hold but the values do not. Which trials are the samples is also open (the legend gives only neuron counts); readings are (g) pseudo- population trials (64) and (h) each mouse's own trials, pooled by group.
- fig2f-latency
In the 30 Hz data OB-p neurons reach their first calcium peak after odor onset sooner than mPFC-p neurons; OB-p and untagged neurons do not differ.
OB-p neurons exhibited shorter latencies to reach the first peak of calcium activity following odor onset compared to mPFC-p neurons
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 3; Figure 2FHow it will be judged
Fast (30 Hz) dataset. Peaks found on normalized fluorescence per trial with the Methods' minimum criteria (width 100 ms, distance 1 s, height 0.5, prominence 0.25), read as scipy.signal.find_peaks arguments, ignoring the first and last second of each trial; latency is the time of the first peak after odor onset. Whether a latency is per trial (all trials with a peak) or per neuron (median over trials) is open; the sizes of the p- values suggest per trial, and both are run, (a) per trial and (b) per neuron. Two-sided Mann-Whitney U tests, Bonferroni corrected over the 3 pairs. Reproduced when OB-p is shorter than mPFC-p with corrected p < 0.05 under one reading, and, under that reading, OB-p against untagged has corrected p >= 0.05 (the statement's claim of absence, on the same test); the "= 1" is not required.
- fig2g-early-peaks
OB-p neurons have more calcium peaks at early time points after odor onset than mPFC-p neurons.
OB-p neurons also had more peaks during earlier time points than mPFC-p neurons
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 3; Figure 2G and Figure S4A to S4CHow it will be judged
Peaks as in fig2f-latency, counted per trial in time bins aligned to odor onset and averaged per neuron. The paper gives no bin width, no window for "earlier" and no test. Readings of "earlier" are (a) 0 to 2 s (the odor pulse) and (b) 0 to 3 s (the early epoch of Figure 1G). Reproduced when the mean number of peaks per trial in that window is higher for OB-p than mPFC-p under both readings; the counts are reported. The OASIS deconvolved- event version (Figure S4) is shown as a variant.
- fig2h-identity-decoding-5s
Odor identity decoded from 5 s windows reaches similar accuracy in size-matched OB- p and mPFC-p pseudo-populations, above chance and slightly below untagged.
both OB-p and mPFC-p pseudo-populations reached similar accuracy levels, significantly above chance but slightly lower than the untagged population ( Figure 2H ; two-sided Mann-Whitney U test, p = 0.398)
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 4; Figure 2HHow it will be judged
A claim of absence for OB-p against mPFC-p, on the paper's test. Linear SVM, 8 odors, mean normalized fluorescence per cell in the window, 5-fold cross-validation, 75- cell pseudo-populations, 200 train/test repeats (Figure 2H legend, right). The window is 0 to 5 s in the legend and 1 to 5 s in the Methods; both are run, (a) 0 to 5 s and (b) 1 to 5 s. Reproduced when the two-sided Mann-Whitney U test of the 200 OB-p against the 200 mPFC-p accuracies has p >= 0.05; not reproduced when p < 0.05. The mean accuracies are reported against chance (12.5%) and against untagged. If the two windows disagree the claim is methods_underspecified.
- fig2i-identity-decoding-early
In an early window (0 to 2 s) OB-p pseudo-populations decode odor identity better than mPFC-p pseudo-populations.
we found that OB-p pseudo-populations outperformed mPFC-p pseudo-populations ( Figure 2I , two-sided Mann-Whitney U test with p = 2.15 × 10⁻ 30 )
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 5; Figure 2IHow it will be judged
As fig2h-identity-decoding-5s with a 0 to 2 s window. Reproduced when OB-p accuracy exceeds mPFC-p with two-sided Mann-Whitney U p < 0.05 over the 200 repeats.
- fig2j-identity-decoding-peaks-earlier
With a sliding window, odor identity decoding accuracy peaks earlier for OB-p than for mPFC-p pseudo-populations.
SVM classification using a sliding window for odor responses revealed that decoding accuracy for OB-p neurons peaked earlier than for mPFC-p neurons
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 5; Figure 2JHow it will be judged
Linear SVM on mean normalized fluorescence in consecutive windows across the trial, trained and tested on the same window, 75-cell pseudo-populations, 100 repeats. The window is 250 ms in the Figure 2J legend and 500 ms in the Methods; at 4.53 Hz a 250 ms window holds one or two frames. Both are run, (a) 250 ms and (b) 500 ms. No test is given. Reproduced when the time of the peak of the mean accuracy curve is earlier for OB-p than for mPFC-p under both readings; the times and the 95% intervals of the peak time over repeats are reported.
- fig2k-mpfc-late-identity
Late in the trial, mPFC-p pseudo-populations keep low but above-chance identity decoding that exceeds OB-p decoding.
mPFC-p pseudo-populations maintained low but above-chance identity decoding that exceeded OB-p decoding later in the trial
Results, OB-p and mPFC-p neurons exhibit broadly similar odor response properties, but differ in dynamics, paragraph 5; Figure 2KHow it will be judged
From the same-window accuracies of fig2j-identity-decoding-peaks-earlier (the diagonal of Figure 2K). "Later in the trial" is not bounded; it is read as after odor offset, (a) 2 to 15 s and (b) 5 to 15 s after onset. Reproduced when, under both readings, the mPFC-p mean accuracy over the window exceeds 12.5% and exceeds the OB-p mean, with repeats as samples (one-sample one-sided t-test against 12.5% and two-sided Mann-Whitney U, p < 0.05, the tests the paper uses for decoding elsewhere).
- fig3bc-concentration-tuning
Similar fractions of OB-p and mPFC-p neurons are tuned to odor identity, but OB-p neurons more often show significant effects of concentration and identity by concentration interactions, which are rare in mPFC-p neurons.
Across time, similar fractions of OB-p and mPFC-p neurons were significantly tuned to odor identity ( Figure 3B ) ... By contrast, OB-p neurons were more likely to show significant main effects of concentration, as well as identity × concentration interactions
Results, Odor concentration is robustly encoded in OB-p but not mPFC-p neurons, paragraph 2; Figure 3B and 3CHow it will be judged
Concentration dataset. Per neuron, a two-way ANOVA (identity, 3 odors, by concentration, 3 levels, with interaction) on mean normalized fluorescence in the first 5 s of odor presentation over trials (Figure 3B legend), p < 0.05 per term; the fraction of neurons significant per term per group. The time-resolved version (Figure 3C) uses sliding windows of unstated width, read as (a) 500 ms and (b) 1 s steps of the same width. The paper gives no test of the fractions. Reproduced when, in the 0 to 5 s analysis, the OB-p fractions for concentration and for interaction both exceed the mPFC-p fractions, and the OB-p minus mPFC-p difference for identity is smaller in absolute size than that for concentration; the time courses are reported.
- fig3de-concentration-decoding
OB-p pseudo-populations decode odor concentration more accurately and earlier than mPFC-p and untagged pseudo-populations, while mPFC-p decoding stays close to chance.
OB-p pseudo-populations reached higher concentration classification accuracy and did so earlier in the trial than both mPFC-p and untagged pseudo-populations, whereas mPFC-p ensembles supported only weak concentration decoding that remained close to chance
Results, Odor concentration is robustly encoded in OB-p but not mPFC-p neurons, paragraph 3; Figure 3D and 3EHow it will be judged
Per odor, a linear SVM decoding the 3 concentrations from mean normalized fluorescence in sliding windows (250 ms in the Figure 3D legend, 500 ms in Figure 3E and the Methods; both run), 75-cell pseudo-populations, 100 repeats, accuracies averaged over the 3 odors; chance 33.3%. No test is given. Reproduced when the peak mean accuracy of OB-p exceeds those of mPFC-p and untagged, the OB-p peak time is earlier than both, and the mPFC-p peak mean accuracy is below the midpoint between chance and the OB-p peak, under both window readings. The midpoint is a reading of "close to chance", recorded as such; the paper gives no bound.
- figs5g-pairwise-concentration
Pairwise concentration decoding (low against medium, medium against high, low against high) is strongest and most sustained in OB-p ensembles, intermediate in untagged neurons and weakest in mPFC-p neurons.
concentration decoding was again strongest and most sustained in OB-p ensembles, intermediate in untagged neurons, and weakest in mPFC-p neurons
Results, Odor concentration is robustly encoded in OB-p but not mPFC-p neurons, paragraph 3; Figure S5GHow it will be judged
As fig3de-concentration-decoding, with binary SVMs for each pair of concentrations per odor (chance 50%), averaged over odors, same-window accuracies. Read as the mean same-window accuracy over 0 to 15 s after onset per group. Reproduced when the order OB-p > untagged > mPFC-p holds for all three pairs under both window readings.
- figs5-sparseness-concentration
Across the three concentrations, response sparseness and the fraction of active cells change only modestly in all three populations.
overall response sparseness and the fraction of active cells changed only modestly in all three populations ( Figure S5 )
Results, Odor concentration is robustly encoded in OB-p but not mPFC-p neurons, paragraph 2; Figure S5A to S5DHow it will be judged
Per concentration (0.1%, 1%, 10% in the Figure S5 legend), the percentage of active cells (binarized, 0 to 5 s) and lifetime sparseness per neuron over the 3 odors, population sparseness per subject (Figure S5D). The paper gives no test and no bound for "modestly"; the values per population and concentration are reported and the claim is judged only on whether no population shows a monotonic change of more than half its value at 0.1% across the range, a reading of "only modestly" recorded as such.
- fig4c-behavior-learning
Go/NoGo performance improves across training days, with no difference between OB- p and mPFC-p labelled mice, reaching 73.7% correct trials on day 3.
Behavioral performance improved as training progressed, with no significant difference between OB-p and mPFC-p labelled mice ... with, on average, 73.7% correct trials (hit and correct rejection) on day 3
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 2; Figure 4CHow it will be judged
Per mouse and task day (1 to 3), percent correct = (hits + correct rejections) / scored CS trials, with hits as relabeled in the Methods (reward within 10 s after the delay), excluding bad trials. A one-way repeated-measures ANOVA on days, and a mixed ANOVA (days within, projection between). Reproduced when F and the day 3 mean are within tolerance, the day effect has p < 0.05 in both ANOVAs and projection and interaction have p >= 0.05. The Methods give the 10 s relabeling for neural analyses, so whether it applies to behavioral accuracy is open; readings are (a) hits as scored (lick in the reward window) and (b) hits as relabeled. Whether the NWB outcome labels already apply the relabeling is the author's to check. Reproduced when the conditions hold under one reading; both are reported.
- fig4c-anticipatory-licking
Anticipatory licking in response to the rewarded odorants increases across training.
an increase in anticipatory licking in response to the rewarded odorants
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 2; Figure 4C right and Figure S6How it will be judged
Lick rate on CS+ trials before the reward window opens. "Anticipatory" is read as (a) the delay, 2 to 5 s after odor onset, and (b) odor and delay, 0 to 5 s. The paper gives no test. Reproduced when the mean over mice of the CS+ lick rate is higher on day 3 than on day 1 under both readings; per-mouse values are reported.
- fig4def-identity-decoding-gng
In expert mice (day 3), OB-p pseudo-populations decode odor identity above chance in all four task phases, peaking early (delay) and declining later, whereas mPFC-p pseudo-populations are near chance early and improve later in the trial.
odor identity decoding in OB-p pseudo-populations was greater than chance across all 4 task phases ... Decoding accuracy peaked during early trial phases (delay phase after odor exposure), but declined later in the trial (action) ... In contrast, mPFC-p pseudo-populations performed near chance during early trial phases, but decoding accuracy increased later in the trial
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 3; Figure 4D, 4E and 4FHow it will be judged
Day 3, 4 odors (chance 25%), linear SVM per task phase (trained and tested on the same phase), 85-cell pseudo-populations from 4 OB-p and 3 mPFC-p mice, 100 repeats; aggregate readings (a) mean and (b) maximum over the phase. One-sample one-sided t-tests against 25% over the 100 accuracies, Bonferroni over 4 phases. Reproduced when, under one aggregate reading, OB-p is above chance in all 4 phases with its highest mean accuracy in the delay phase and lower in act-late than in delay, and mPFC-p has its two lowest mean accuracies in odor and delay and its highest in act-early or act-late. The Figure 4F legend's OB-p against mPFC-p differences in every phase are reported with two-sided Mann-Whitney U tests (the paper's test for such comparisons in Figure 4H).
- fig4g-valence-any
Generalized odor valence can be decoded above chance from a general PCx (any-p) pseudo-population in all task phases in expert mice.
We found that odor valence could be decoded with high accuracy from a general PCx pseudo-population across all phases
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 4; Figure 4G and Figure S8HHow it will be judged
Day 3. Generalized valence (Methods) trains on one CS+/CS- pair and tests on the other, all 4 permutations, accuracy averaged over them; any-p pseudo-populations of 85 cells sampled from tagged and untagged cells of all 7 mice, 100 repeats, per task phase; aggregate readings (a) mean and (b) maximum. One-sample one-sided t-tests against 50%, Bonferroni over 4 phases. Reproduced when all 4 phases have corrected p < 0.05 under one reading.
- fig4h-valence-timing
In expert mice OB-p pseudo-populations encode generalized valence in the odor and delay phases but fall to chance in the action phases, while mPFC-p pseudo-populations encode valence in the later but not the earliest phase; the two differ in all four phases.
OB-p pseudo-populations accurately encoded odor valence during early phases of the trial (odor exposure, delay), but that decoding accuracy decreased to chance level later in the trial (action) ... In contrast, mPFC-p pseudo-populations encoded odor valence during later but not early phases of the trial
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 5; Figure 4HHow it will be judged
As fig4g-valence-any with OB-p and mPFC-p pseudo-populations. Reproduced when, under one aggregate reading, OB-p is above 50% (corrected p < 0.05) in odor and delay and not in act-late; mPFC-p is not above 50% in odor and is above in delay, act-early and act- late; and the two-sided Mann-Whitney U test of OB-p against mPFC-p accuracies has corrected p < 0.05 in all four phases. Partially reproduced when the pattern of above-chance phases holds and the between-group tests hold in two or three phases. The text's act-early OB-p result (p = 0.00141, above chance) is reported alongside.
- fig4ij-valence-correct-trials
Restricted to correct trials, OB-p pseudo-populations encode valence well early and much less later, mPFC-p pseudo-populations encode it more accurately late than early, and the two differ in all four phases.
OB-p pseudo-populations accurately encoded odor valence during early phases of the trial but much less so later ... mPFC-p pseudo-populations encoded odor valence late in correct trials more accurately than early ... For all four task phases, these projection pseudo-populations performed differently from one another at decoding valence for correct trials
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraphs 6 and 7; Figure 4I and 4JHow it will be judged
Day 3, correct trials only (hits and correct rejections). Which valence scheme Figure 4I and 4J use is open, since the legend says generalized valence in correct trials is in Figure S8E and S8G; the reading is valence (unconstrained) for correct trials (Methods, the fifth classifier), with the generalized scheme as the subject of figs8-generalized-valence-correct. 85 cells, 100 repeats, aggregate readings (a) mean and (b) maximum. Reproduced when, under one reading, OB-p mean accuracy is higher in odor and delay than in act-early and act-late, mPFC-p is higher in act-early and act-late than in odor and delay with odor not above 50% (corrected p >= 0.05), and OB-p against mPFC-p differs (two-sided Mann-Whitney U, Bonferroni, p < 0.05) in all four phases.
- figs8-generalized-valence-correct
Generalized valence decoding in correct trials differs between OB-p and mPFC-p pseudo-populations in the odor and act-early phases but not in the delay and act-late phases.
We also observed significant differences with generalized valence decoding performance in correct trials between projection pseudo-populations during odor and act-early phases
Results, Differential encoding of odor valence in OB-p and mPFC-p neurons, paragraph 7; Figure S8E, S8G and S8KHow it will be judged
Day 3, correct trials, generalized valence (4 permutations), 85 cells, 100 repeats, aggregate readings (a) mean and (b) maximum. Reproduced when, under one reading, the two-sided Mann-Whitney U test of OB-p against mPFC-p has corrected p < 0.05 in odor and act-early and corrected p >= 0.05 in delay and act-late; the direction in each phase is reported.
- tracked-neuron-counts
Longitudinal tracking across the three Go/NoGo sessions follows 46 OB-p and 41 mPFC-p neurons in 4 OB-p and 3 mPFC-p tagged mice.
we longitudinally tracked identified neurons across three Go/NoGo sessions and followed 46 OB-p and 41 mPFC-p neurons in 4 OB-p–tagged and 3 mPFC-p–tagged mice
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 1; Methods, Longitudinal trackingHow it will be judged
Cells with one tracking identifier present in all four sessions (day 0 to 3) and red in at least 3 of them (code criterion min_3_sessions), counted per projection. Reproduced when both counts are within tolerance; the counts under the code's other criteria (min_1_session, first_4_sessions, all_sessions) are reported.
- figs10d-slope-no-difference
Tracked OB-p and mPFC-p neurons do not differ in the average slope of their session-averaged normalized fluorescence across the three Go/NoGo days.
there was no significant difference in the average slope between OB-p and mPFC-p neurons
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 2; Figure S10DHow it will be judged
A claim of absence, on the paper's test. For each tracked neuron and each combination of odor valence (CS+, CS-) and task phase (4), the mean normalized fluorescence over the phase and over that valence's trials on each of days 1 to 3, and the least-squares slope over days; a mixed ANOVA with the combination (8 levels) within neurons and projection between. Reproduced when the projection effect has p >= 0.05; not reproduced when p < 0.05. The interaction is reported against p = 0.0031.
- figs10e-sparseness-slope-no-difference
Changes in lifetime sparseness across sessions show no clear effect of projection target.
We similarly quantified changes in lifetime sparseness across sessions and did not observe a clear effect of projection target
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 2; Figure S10EHow it will be judged
A claim of absence. Per tracked neuron, task phase and day, lifetime sparseness over the 4 odors of binarized activity in that phase; slope over days 1 to 3; a mixed ANOVA with phase within and projection between. Reproduced when the projection effect has p >= 0.05; not reproduced when p < 0.05.
- fig5c-cv-higher-obp
Tracked OB-p neurons show a higher coefficient of variation of session-averaged fluorescence across days than mPFC-p neurons.
OB-p neurons typically showed higher fluorescence CV across days than mPFC-p neurons ( Figure 5C, D ). A mixed ANOVA on CV within combinations of odors and task phases revealed a significant effect of projection target (p = 0.0127)
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 3; Figure 5CHow it will be judged
Per tracked neuron and combination, the coefficient of variation (SD over mean) of the session-averaged normalized fluorescence over days 1 to 3; a mixed ANOVA with the combination within neurons and projection between. Open details are the combination, read as (a) valence (CS+, CS-, as plotted in Figure 5C) by phase, 8 levels, and (b) odor (4) by phase, 16 levels; and the treatment of means near zero or negative (normalized fluorescence is centred near zero), read as (c) SD over the absolute mean and (d) SD over the mean. Reproduced when, under every combination of readings, the projection effect has p < 0.05 with OB-p higher (by median CV); when readings disagree, methods_underspecified. The per-condition Mann-Whitney U tests with Benjamini-Hochberg correction (Figure S10F) are reported.
- fig5e-transferability
Identity decoding transfers across sessions comparably in tracked OB-p and mPFC-p ensembles, whereas generalized valence decoding transfers better in mPFC-p ensembles, especially across distal sessions.
When decoding odor identity, transferability across dissimilar sessions was comparable between OB-p and mPFC-p ensembles ... whereas mPFC-p ensembles maintained higher valence decoding accuracy, particularly for generalized valence across distal sessions
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 4; Figure 5E and Figure S11How it will be judged
Per mouse, a linear SVM on its tracked projection neurons, trained on one (day, phase) and tested on another, for days 1 to 3 and the 4 phases (Figure 5D); same-phase and adjacent-phase pairs only; transfers grouped as within session, proximal (consecutive days) and distal (day 1 and day 3). Open details, which could change the outcome. The samples of the Mann-Whitney test, read as (a) every qualifying accuracy entry of every mouse and (b) per mouse means; the folds for train and test within one session, read as 5- fold stratified cross-validation; aggregate readings mean and maximum. Two-sided Mann- Whitney U tests of OB-p against mPFC-p entries per group, Bonferroni over the 3 groups. Reproduced when, under one reading, identity has corrected p >= 0.05 for proximal and distal, and generalized valence has mPFC-p higher with corrected p < 0.05 for distal.
- fig5f-stability
The temporal structure of valence coding across sessions is more stable in tracked mPFC-p than OB-p ensembles, while odor identity coding is not.
These analyses revealed that the temporal structure of valence coding was consistently more stable in mPFC-p than in OB-p ensembles
Results, Longitudinal tracking reveals greater stability of valence encoding in feedforward projection neurons, paragraph 5; Figure 5F and Figure S12BHow it will be judged
From the per-mouse transfer matrices of fig5e-transferability, each 4 by 4 transfer block (train day, test day) flattened, and Pearson correlations between pairs of blocks (Methods, Decoding with tracked neurons). Open details are which blocks are paired, read as (a) all pairs of the 9 blocks and (b) pairs sharing the training day; and the samples, read as (c) every correlation of every mouse and (d) per mouse means. Two-sided Mann-Whitney U tests of OB-p against mPFC-p correlations, Bonferroni over the 4 schemes. Reproduced when, under one combination of readings, valence, generalized valence and correct-trial valence each have mPFC-p higher with corrected p < 0.05 and odor identity has corrected p >= 0.05.
- discussion-differences-grow
The differences in timing of identity and valence encoding between OB-p and mPFC- p ensembles increase over the course of learning.
Notably, these differences increased over the course of learning
Discussion, Functional specializations of projection neurons align with those of their target areas, paragraph 1; Figure S7A and Figure S9How it will be judged
As in the Figure S7A and S9 legends. Per projection and task phase, accuracies of 50 size-matched pseudo-populations per day are sampled and the slope of accuracy over days 1 to 3 is fitted, 1000 times; a null from day-shuffled data; two-sided Mann-Whitney U tests with Bonferroni correction of observed against null slopes, for identity and generalized valence. The paper gives no test of the difference itself; the claim is read as reproduced when the slopes are significantly positive (corrected p < 0.05, above the null) for OB-p in the odor or delay phase and for mPFC-p in an action phase, for identity or generalized valence. Aggregate readings mean and maximum.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none