Claims 28
From the paper Experience-dependent contextual codes in the hippocampus · claims file · checked 2026-10-22: ready
- cells-per-session
Each imaging session recorded between 161 and 3,657 putative CA1 pyramidal cells.
In each session, we were able to simultaneously image up to several thousands (range of 161–3,657) of putative CA1 pyramidal cells
Results, Two-photon imaging of CA1 cells in morphed VR environments, paragraph 1; Methods, Two-photon imaging; Extended Data Fig. 1a,bHow it will be judged
Count of cells (Suite2P regions kept as cells) per imaging session (Methods, Two-photon imaging). Which mice the range covers is not stated; Extended Data Fig. 1a,b plot R and F mice only, while the Methods speak of the whole study. Readings (a) every imaging session of R and F mice, (b) every imaging session of R, F and FD mice. Reproduced when, under a reading, the minimum and maximum fall within the range 161 to 3,657 widened by 10% of each bound.
- session-and-trial-counts
The analyses of sessions 8 to N use 24 rare morph and 21 frequent morph sessions, holding 2,610 and 1,986 trials; the four decision mice contribute 2,460 trials.
How it will be judged
Sessions 8 and later per mouse (M1); trials are all imaged trials of those sessions. For the decision mice the paper does not say which sessions; read as (a) every imaged session and (b) sessions 8 and later. Reproduced when every count is within tolerance under a reading.
- training-intermediate-fraction
Before session 8, rare morph mice had fewer than 4% of trials with intermediate morph values, while frequent morph mice had 60%.
mice experienced very few trials with intermediate morph values (shared S ~ = 0.25 − 0.75 ) during the first 7 training sessions (<4% of trials with intermediate morph values before session 8 ... mice were frequently exposed to trials with intermediate morph values (60% of trials with intermediate morph values before session 8
Results, Two-photon imaging of CA1 cells in morphed VR environments, paragraph 4; Figure 1d,eHow it will be judged
Per mouse, the fraction of all trials run in sessions 1 to 7 (imaged and not, M7) with S-tilde 0.25, 0.5 or 0.75. The paper does not say whether the fraction is pooled over mice or per mouse; readings (a) pooled over trials of the group, (b) every mouse. Reproduced when, under a reading, the rare morph fraction is below 4% and the frequent morph fraction is within 0.06 of 60%.
- place-cell-fractions
Place cells tiled the track in both conditions; 15,829 rare morph place cells (43% of recorded cells; 23% S-tilde = 0 and 21% S-tilde = 1 place cells) and 8,933 frequent morph place cells (30%; 11% and 10%).
rare = 15,829 place cells across all S ~ values, 43% of recorded population, 23% of cells S ~ = 0 place cells, 21% of cells S ~ = 1 place cells; frequent = 8,933 place cells across all S ~ values, 30% of recorded population, 11% of cells S ~ = 0 place cells, 10% of cells S ~ = 1 place cells
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 1How it will be judged
Methods, Place cell identification. Per session and separately for each S-tilde value, Skaggs spatial information of each cell over the 45 bins (M3) from the trial-averaged rate map and the occupancy of those trials; the code subtracts the global minimum of the map and adds 0.001 before the log (PlaceCellAnalysis.spatial_info), followed. Null from 1,000 shuffles, each circularly shifting the activity against position within every trial by a uniform random amount (utilities.make_pos_bin_trial_matrices, perm=True); a place cell has p = fraction of shuffles with SI at least the real SI at most 0.05. A cell is a place cell "across all S-tilde" if it is one for any S-tilde. Sessions 8 to N (M1); signal M2. Open details, (1) speed exclusion (M3 readings a and b); (2) the denominator, read as (a) all cells of the condition's sessions pooled, (b) the mean of per-session fractions. Reproduced when all eight values are within tolerance under one combination; partially reproduced when the counts or the fractions alone are.
- edf4a-fewer-place-cells-frequent
A smaller proportion of recorded cells were place cells in the frequent morph than in the rare morph condition.
The proportion of recorded cells that were classified as place cells was smaller in the frequent morph compared to the rare morph condition
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 1; Extended Data Fig. 4aHow it will be judged
The place cells of place-cell-fractions, under each of its readings. The paper reports no test. Reproduced when the frequent morph fraction (any S-tilde, and for S-tilde = 0 and = 1) is below the rare morph fraction under every reading; the per-mouse fractions are shown.
- edf4a-overlap-above-chance
The fraction of cells that are place cells in both S-tilde = 0 and S-tilde = 1 trials (6.07% rare, 1.98% frequent) slightly exceeds the overlap expected of independent populations (4.78% and 1.06%), so the two extreme maps are largely independent (global remapping).
Specifically, 6.07% and 1.98% of recorded cells had significant spatial information in both environments for rare morph and frequent morph sessions, respectively (chance 4.78% and 1.06%, respectively; one-sample two-sided t -test of actual minus chance overlap, t =−4.87, P = 1.37×10 −5 , N = 45 sessions pooled across all mice)
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 2; Extended Data Fig. 4aHow it will be judged
Place cells as in place-cell-fractions. Per session, actual overlap is the fraction of all cells that are place cells for both S-tilde = 0 and = 1; chance is the product of the two fractions (Results; code PlotAllPlaceCells, randovlp and actlovlp). The group values in the code are pooled; read as (a) pooled over cells and (b) the mean over sessions. Test, as the code runs it, scipy ttest_1samp of chance minus actual over all sessions of both groups (the order that gives the paper's negative t). Reproduced when the test gives p < 0.05 with actual above chance and the four values are within tolerance under one reading; partially reproduced when only the test holds. The 2020 code ran it on 47 sessions; the attempt reports its count.
- edf4bc-overlap-cells-at-ends-and-reward
Cells with place fields in both extreme environments (2,219 rare, 594 frequent) tended to cluster at the end points of the track and in the reward zone.
Cells with place fields in both extreme environments tended to cluster at the end points of the track and the reward zone
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 2; Extended Data Fig. 4b,cHow it will be judged
Cells of edf4a-overlap-above-chance that are place cells in both. Location is the position bin of peak trial-averaged activity in S-tilde = 0 trials and in S-tilde = 1 trials (Extended Data Fig. 4b,c). The paper shows this only as scatter plots; it gives no region or test. The reward zone is 250 to 400 cm (Methods, Morphed environments); "end points" is read as (a) the first and last bin and (b) the first and last 50 cm. The reference is (a) the share of the track those regions cover and (b) the peak locations of all S-tilde = 0 place cells. Reproduced when, in both conditions, the share of overlap cells with peaks in the end points or reward zone exceeds the reference under every combination of readings; the counts are judged by their tolerances.
- fig2eh-place-cell-similarity-differs
Rare morph place cells keep their map up to a threshold morph value (S about 0.25 to 0.5) and then switch coherently, while frequent morph place cells change gradually; the difference shows in the place-cell similarity matrices across S-tilde values.
In the rare morph condition, place cells maintained their representation until a threshold morph value ( S ≈ 0.25–0.50) and then coherently switched to a different representation ... This difference between conditions was also apparent in the normalized place cell population similarity matrices ( P < 0.0011, two-sided permutation test using animal identities (IDs), n = 12 mice)
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 3; Figure 2e-hHow it will be judged
Methods, Place cell similarity matrices. Cells that are place cells for any S-tilde (place-cell-fractions), pooled over sessions 8 to N and mice of a condition. Per cell, the five trial-averaged rate maps (one per S-tilde, smoothed per M5), rows divided by their l2 norm, give a 5 x 5 cosine matrix; matrices are averaged over the cells of a condition and subtracted (rare minus frequent). For the z-scored version each cell's matrix is z-scored over its off-diagonal entries (Methods; the 2020 code z-scores jointly over all cells of a session, a variant). Mouse permutation test M12 on each entry, all cells of a mouse moving together, with the 6 rare and 6 frequent mice of the paper (the 2020 code counts R6 as frequent, mi<5, a variant). Reproduced when at least one off-diagonal entry of the z-scored difference lies outside the range of the 923 permuted differences, and at least one entry of the raw difference lies outside their 2.5 to 97.5 percentiles. The entries that do, with their signs, are reported.
- edf5-umap-rare-frequent-separable
In a nonlinear embedding (UMAP) of the morph-by-position activity of task-modulated cells, cells from rare morph and frequent morph mice can be told apart.
Rare and frequent morph cells were also separable using this approach.
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 4; Extended Data Fig. 5cHow it will be judged
Methods, UMAP for dimensionality reduction. The Methods do not name the sessions; read as sessions 8 to N, the sessions of the remapping analyses (Results, Prior experience..., paragraph 1). Cells kept when the Euclidean distance between their mean S-tilde = 0 and S-tilde = 1 rate maps exceeds 99.9% of 1,000 shuffles of the S-tilde labels of those trials (the Methods report 18,230 cells kept, 150 of them dropped between the manifolds; the attempt reports its counts); maps binned by S into ten bins (edges not stated; read as ten equal bins over S from -0.1 to 1.1, a standard choice) and flattened (450 values per cell); UMAP (n_neighbors 100, correlation, min_dist 0.1, 3 dimensions; the code's min_dist 0.01 is a variant); DBSCAN into two manifolds (the code's min_samples 200, default eps), points outside both dropped; k-means with k = 10 in each; per cluster a new 10-dimensional UMAP (n_neighbors 20, min_dist 0.1, correlation); logistic regression with ridge penalty and balanced class weights (code default C = 1, 10-fold cross-validation, permutation_test_score) classifies rare against frequent cells; significant when accuracy exceeds all label shuffles. The number of shuffles disagrees, 100 in the Methods and 1,000 in the Extended Data Fig. 5c legend; both are run, and a cluster is significant under that reading when it exceeds all shuffles. The paper does not say how many clusters make the two "separable"; readings (a) most of the 20 clusters significant, (b) at least one. UMAP and k-means seeds are a standard choice. The outcome is the one all readings give.
- edf6-nmf-confirms-remapping
Non-negative matrix factorization of single-cell trial-by-trial similarity matrices confirms discrete remapping in rare morph sessions and gradual remapping in frequent morph sessions.
These findings were also confirmed using a second unsupervised approach ... non-negative matrix factorization
Results, Prior experience of stimulus frequency determines CA1 place cell remapping, paragraph 4; Extended Data Fig. 6How it will be judged
Extended Data Fig. 6a legend. Per session 8 to N, each cell's trial-by-trial cosine similarity matrix (Methods, Single-cell trial-by-trial similarity matrices; maps smoothed per M5), upper triangle flattened, stacked over cells, factorised with rank 3; each factor reshaped into a trial-by-trial prototype and sorted by S. The paper judges the factors by eye and gives no measure. Read as follows (a reading, not from the paper, the same measure as fig3ab-block-versus-gradual-similarity). For each factor, average its prototype over S-tilde blocks (5 x 5, diagonal left out); for the factors whose largest block is (0, 0) or (1, 1), the step share of that extreme's row is the largest fall between neighbouring S-tilde values divided by the whole fall to the other extreme. Per session the mean over those factors, per mouse the mean over sessions. Reproduced when the across-mouse mean step share is higher for rare than for frequent morph mice and test M12 (2.5 and 97.5 percentiles) finds the difference. The 2020 code bins all sessions of a mouse into ten S bins with ranks 2 and 3 (SingleCellNMF), a variant.
- fig3ab-block-versus-gradual-similarity
Single-session population trial-by-trial similarity matrices show two blocks in the rare morph condition and a gradual transition between the extremes in the frequent morph condition.
For the rare morph condition, a strong block diagonal structure in the trial-by-trial similarity matrices suggested clustering of morph values into two representations
Results, Population similarity analyses reveal distinct remapping patterns visible at the level of single trials, paragraphs 1 and 2; Figure 3a,b; Extended Data Figs. 7 and 8How it will be judged
Methods, Population trial-by-trial similarity matrices. Per session, all cells' single trial rate maps (M2, M3, M5) concatenated per trial, rows l2-normalised, cosine matrix C; rows and columns binned by S-tilde, diagonal left out; binned matrices averaged over sessions per mouse. The paper judges the single sessions by eye and gives no measure of a block or of a gradual transition; the group comparison it tests is fig3f-zscored-difference-at-025, on which this claim depends. Read as follows (a reading, not from the paper). In the S-tilde = 0 row and in the S-tilde = 1 row of a mouse's binned matrix, the step share is the largest fall in similarity between neighbouring S-tilde values divided by the whole fall from the row's own extreme to the other extreme; two blocks give a share near 1, an even gradient one near 0.25. A mouse's value is the mean of its two rows. Reproduced when the across-mouse mean step share is higher for rare than for frequent morph mice and test M12 (2.5 and 97.5 percentiles) finds the difference; the per-mouse shares are shown.
- edf7-checkerboard-sessions
In six named rare morph sessions, trials near S-tilde = 0.25 randomly take one extreme representation or the other, seen as checkerboard bands in the sorted similarity matrices.
checker-board dark bands were observed in the similarity matrices near S ~ = 0.25 , as if the matrix was missorted for R1 session 8, R2 session 13, R3 session 9, R4 session 11, R5 session 8 and R6 session 10
Results, Population similarity analyses reveal distinct remapping patterns visible at the level of single trials, paragraph 1; Extended Data Fig. 7How it will be judged
The paper shows this by eye only. Read with the paper's own measure of which extreme a trial resembles, the whole-trial SF (M9). Reproduced when, in each of the six sessions named, the S-tilde = 0.25 trials include at least one trial with SF above 0.5 and at least one below 0.5. Session numbers follow the paper's numbering; how they map onto the dataset is the author's to show.
- fig3f-zscored-difference-at-025
After z-scoring within session, trial-by-trial population similarity differs significantly between rare and frequent morph mice for S-tilde = 0.25 trials.
As predicted, group differences in trial-by-trial similarity were significant for S ~ = 0.25 trials after z -scoring within session similarity ( P < 0.05 two-sided mouse permutation test, n = 12 mice)
Results, Population similarity analyses reveal distinct remapping patterns visible at the level of single trials, paragraph 3; Figure 3c-fHow it will be judged
Methods, Population trial-by-trial similarity matrices. Per session 8 to N, C as in fig3ab-block-versus-gradual-similarity; binned into the five S-tilde values (code PlotPopSimMats) as means over off-diagonal entries; the 25 binned entries z-scored within the session (Methods; the 2026 code z-scores the mouse means, a variant); averaged over sessions per mouse; rare minus frequent across-mouse means; permutation test M12 at the 2.5 and 97.5 percentiles of the 923 permuted differences. Reproduced when at least one entry of the S-tilde = 0.25 row is significant; the significant entries and their signs are reported.
- fig4-sf-stable-along-track
Position-wise SF shows stable clustered representations along the whole track in rare morph mice and a spectrum in frequent morph mice; remapping happens soon after the trial starts and stays as the mouse runs.
Plotting SF as a function of the position of the animal along the VR track revealed a clustering of stable representations across the entire length of the track for the rare morph condition ... Together, these results indicate that remapping rapidly occurred after the start of a new trial and remained stable as the animal traversed the virtual track.
Results, CA1 representations show stable discrimination of morph values along the length of the track, paragraphs 1 and 2; Figure 4b-eHow it will be judged
Position-wise SF (M9) for every trial of sessions 8 to N, averaged per S-tilde within session, then over sessions per mouse, then across mice with s.e.m. (Fig. 4d,e legends). The paper reads the curves by eye. Read as reproduced when (1) in both conditions the mean SF of S-tilde = 0 and of S-tilde = 1 differ by more than the sum of their s.e.m. in every bin after the first bin; (2) in rare morph mice the S-tilde 0.75 and 1.0 curves overlap within s.e.m. over most bins (Fig. 4d legend) while the S-tilde 0.25 curve stays nearer the S-tilde 0 than the S-tilde 1 curve; (3) in frequent morph mice the S-tilde 0.25 curve lies between the S-tilde 0 and 0.5 curves over most bins.
- fig6a-dkl-sign
Delta D_KL of each session's Q estimate took the expected sign (negative for rare, positive for frequent) in 41 of 45 sessions and in every session of 9 of 12 mice; across mice, rare -0.88 +- 0.23 and frequent 0.50 +- 0.11 (mean +- s.e.m.), a significant difference.
This Δ D KL measure took the expected sign (rare sessions Δ D KL < 0 and frequent sessions Δ D KL > 0) in 41 out of 45 sessions, and it took the expected sign in every session from 9 out of 12 mice (across animal mean±s.e.m.: rare Δ D KL = −0.88 ± 0.23, frequent Δ D KL = 0.50±0.11; P <0.0011, two-sided animal permutation test, n = 12 mice)
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 2; Figure 6aHow it will be judged
Per session 8 to N, SF (M9), Q (M10) and Delta D_KL (M11) against the ideal rare and frequent posteriors (M6 to M8); per-mouse mean over sessions; test M12. Tolerances are the paper's s.e.m. for the means and 10% for the counts. Reproduced when the test gives the observed difference outside the range of all permutations and all four values are within tolerance; partially reproduced when the test and the signs hold but a value is outside. The code variants (likelihood widths of M8, the last-mouse group prior of M7, signal of M2) are shown and do not decide. The 2020 code printed rare -0.96 +- 0.31 and frequent 0.58 +- 0.14 (SFReg); no code for the 41 of 45 count or the test was found.
- fig6bc-pooled-dkl
Q estimates pooled over all sessions and mice approximate the posterior of the matching condition (Delta D_KL rare -0.44, frequent 0.42).
Combining data from all sessions and all mice gave a robust approximation (Δ D KL ) of the posterior for both the rare morph and frequent morph conditions (rare Δ D KL = −0.44, frequent Δ D KL = 0.42
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 3; Figure 6b,c, left panelsHow it will be judged
Per mouse, the across-session Q estimate is the average of its session Q estimates, renormalised (Fig. 6b legend); the group Q sums the mouse Q estimates and renormalises (Methods, Estimating posterior distributions, "weighting mice equally"); Delta D_KL as M11. An alternative the Methods also allow (Reconstructing prior distributions, "pooled transformed SF values across all sessions for that mouse") is (b) per mouse one Q from all its trials pooled. Reproduced when both values are within tolerance under a reading; partially reproduced when only the signs hold.
- fig6d-prior-reconstruction
Priors reconstructed from each mouse's neural data alone are closer to the ideal prior of the mouse's own condition (Delta D_KL rare -0.46 +- 0.12, frequent 0.46 +- 0.07, mean +- s.e.m. across mice), a significant group difference.
A similar Δ D KL metric showed that the reconstructed priors are closer to the ideal prior from the correct condition than the ideal prior from the incorrect condition ( Fig. 6d ) (across animal mean±s.e.m.: rare Δ D KL = −0.46 ± 0.12, frequent Δ D KL = 0.46±0.07; P <0.0011, animal permutation test, n = 12 mice)
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 4; Figure 6b-dHow it will be judged
Methods, Reconstructing prior distributions. Per mouse, transformed SF values of all sessions 8 to N pooled into one unnormalised Q (M10); R = sum over f-hat_h of Q(f_h | f-hat_h) divided by sum over f-hat_h of the likelihood P(f-hat_h | f_h) / Z(f-hat_h), Z the row sum of the unnormalised Q, then normalised; Delta D_KL = D_KL(P_rare prior || R) - D_KL(P_freq prior || R), in bits. The Methods formula writes Q in the second term; read as R, as the code computes it (SFReg, prior_post_reconstruction). Test M12 on the per-mouse values. Reproduced when the test puts the difference outside all permutations and both means are within tolerance; partially reproduced when only the test and signs hold.
- map-estimate-values
The average MAP estimate of f_h for S-tilde = 0 trials is 11.08 under the rare and 10.64 under the frequent posterior, and for S-tilde = 1 trials 38.13 and 37.30.
average rare morph MAP estimate for S ~ = 0 trials: f h = 11.08; average frequent morph MAP estimate for S ~ = 0 trials: f h = 10.64; average rare morph MAP estimate for S ~ = 1 trials: f h = 38.13; average frequent morph MAP estimate for S ~ = 1 trials: f h = 37.30
Methods, Estimating posterior distributions from population activity (Q estimates)How it will be judged
Argmax over f_h of the group posteriors (M6 to M8) at the f-hat_h of each S-tilde = 0 or 1 trial, averaged over trials. Which trials is not stated; readings (a) all such trials of sessions 8 to N of both groups, (b) of each group's own sessions. Reproduced when all four are within tolerance under a reading. The code variants of M7 and M8 are shown.
- edf9-early-sessions
The rare versus frequent remapping patterns were stable over sessions and emerged early in training (session 3) in most animals.
Of note, these remapping patterns observed in rare morph versus frequent morph conditions were stable over sessions and emerged early in training in most animals
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 5; Extended Data Fig. 9a-iHow it will be judged
Delta D_KL (M9 to M11) of each mouse's session 3 (R6 has none, Extended Data Fig. 9c), and per session for sessions 3 and 8 to N (Extended Data Fig. 9c,d). The ideal posteriors for session 3 are not stated; readings (a) the group posteriors of M8, (b) posteriors from the trials each mouse ran before session 3. "Most animals" is read as more than half of the 11 mice with a session 3. Reproduced when session 3 Delta D_KL has the expected sign (rare negative, frequent positive) in at least 6 of 11 mice under both readings, and in sessions 8 to N the sign of every mouse's per-session values is the same in most of its sessions.
- edf9k-null-models-rare
For rare morph sessions, Q estimates are fit worse by the frequent, uniform, Gaussian and multimodal prior posteriors and by a uniform posterior than by the rare morph posterior.
Furthermore, a comparison of Q estimates to a number of other null model posteriors yielded poor fits to the data
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 5; Extended Data Fig. 9j,kHow it will be judged
Per rare morph session 8 to N, the mean over f-hat_h (M11 rows) of D_KL(P || Q) for the rare posterior and for each null; Wilcoxon signed-rank of null minus rare over sessions (code NullModels4GrumpyPantsReviewer). Null priors not in the Methods, taken from the code (cell 3) on the M6 grid; uniform; Gaussian exp(-(x - 0.5)^2/0.2^2); multimodal, weights 1, 1, 10, 1, 1 at S 0, 0.25, 0.5, 0.75, 1 with exp(-d^2/0.08^2) bumps (in S, through M6); and a constant posterior. Each null posterior with the likelihood of M8. Reproduced when every one of the five tests gives p < 0.05 with the rare posterior closer. The 2020 code printed p = 4.68e-3 for the uniform prior, against 4.7e-4 in the legend.
- edf9l-null-models-frequent
For frequent morph sessions, Q estimates are fit worse by the rare, uniform, Gaussian and multimodal prior posteriors and by a uniform posterior than by the frequent morph posterior.
Wilcoxon signed rank tests compare D KL for the frequent morph condition to D KL for all other posteriors in ( j )
Extended Data Fig. 9l legendHow it will be judged
As edf9k-null-models-rare, for the frequent morph sessions 8 to N against the frequent posterior. Reproduced when all five tests give p < 0.05 with the frequent posterior closer.
- edf10-decision-mice-like-frequent
Frequent morph mice trained to report whether S was above or below 0.5 by where they licked (FD1 to FD4) remapped like frequent morph mice without the task.
The remapping patterns were similar to those observed in frequent morph mice that did not perform this task
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 6; Extended Data Fig. 10How it will be judged
Extended Data Fig. 10f-h. SF, Q and Delta D_KL as in fig6a-dkl-sign and the prior reconstruction as in fig6d-prior-reconstruction, for FD mice; FD sessions are read as in session-and-trial-counts. The comparison is not stated; readings (a) the Q estimate pooled over FD mice has Delta D_KL > 0 (closer to the frequent posterior), (b) each FD mouse's reconstructed prior has Delta D_KL > 0. Reproduced when both readings hold; partially when one.
- edf1-licking-before-reward
Well-trained mice ran steadily down the first half of the track and began licking as they approached the reward.
Once the animals were well trained, they often ran consistently down the first half of the track and began licking as they approached the reward
Methods, Morphed environments, paragraph 1; Extended Data Fig. 1c-gHow it will be judged
Lick rate in the 10 cm bins of M3 per trial, sessions 8 to N. Extended Data Fig. 1g bins trials by reward location (50 cm bins) and plots lick rate against position, with no test. Read as reproduced when, in the across-mouse mean for each reward-location bin, lick rate rises over the 50 cm before the reward location and its peak lies within 50 cm of the reward location, and per mouse the mean lick rate over 0 to 225 cm is below that over 250 to 400 cm. The running part of the statement is shown in no figure (Extended Data Fig. 1 shows licking only) and is not judged; mean speed by position is shown.
- edf2-fh-best-predictor
Of the stimulus features, the horizontal component of wall frequency f_h best predicts the first principal component of the variance across VR scenes.
The horizontal component of wall frequency, f h , is the strongest predictor of the projection onto PC1.
Extended Data Fig. 2g legend; Results, Two-photon imaging of CA1 cells in morphed VR environments, paragraph 3How it will be judged
Extended Data Fig. 2a-g. Uses screenshots of the VR scene every 10 cm for the range of wall, tower and background morph settings, their per-channel FFT power spectra, PCA across settings, and single-predictor linear regressions onto PC1. These are renders of the Unity scenes, not recorded data. Reproduced when the f_h model has the lowest mean squared error of the single-feature models.
- fig7-model-reproduces-remapping
A Hebbian network with radial basis inputs and k-winners-take-all competition, trained on stimuli drawn from the rare or frequent morph prior, gives Q estimates that match the matching posterior.
model-derived SF values and Q estimates were good matches to the appropriate posterior distributions
Results, Simple associative-learning mechanisms can approximate probabilistic inference in neural populations, paragraph 1; Figure 7b-fHow it will be judged
Methods, Hebbian learning model (the parameter list says "For Fig. 4", read as Fig. 7). M = 100 inputs with radial basis tuning exp(-(mu_j - f_h)^2/sigma^2) (the Methods drop the minus sign), sigma^2 = 4.11, centres tiling the stimulus axis; N = 100 outputs, K = 20, c_y = 0.01, sigma_y 0.01 N(0,1), sigma_S 2.74 N(0,1), eta = 0.1, tau = 0.001, W_max = 5, W initial U(0, 0.1). The Methods also list sigma_w 0.05 N(0,1) but place it in no equation; readings (a) added to each weight update, (b) left out; both are run. 1,000 training stimuli drawn from the group prior of M7; 50 models per condition (Fig. 7c legend). Not in the Methods, from the code (KWTA_EmpiricalPrior), testing on the 1,000-point grid of M6, SF centroids from test stimuli below 0.05 and above 0.9 units, the affine map on medians of [-0.1, 0.1] and [0.9, 1.1], then Q and Delta D_KL as M10 and M11. Seeds are a standard choice. Reproduced when the median Delta D_KL of the 50 rare-trained models is below 0 and that of the 50 frequent-trained models above 0 (Fig. 7f). The model uses no recorded activity; it needs only the trial morph values for the prior.
- fig7g-model-prior-reconstruction
Priors reconstructed from the model's Q estimates are closer to the prior of the matching condition, though less closely matched than for the neural data.
These unsupervised reconstructions were quantitatively more similar to the prior of the matching condition, but the match between the reconstruction and the true priors was not as compelling as the match seen in the neural data.
Results, Simple associative-learning mechanisms can approximate probabilistic inference in neural populations, paragraph 1; Figure 7gHow it will be judged
Models as in fig7-model-reproduces-remapping; R per model as in fig6d-prior-reconstruction, averaged over the 50 models (Fig. 7g legend, "Delta D_KL is for the averaged reconstructed prior"), in bits (the code's nats are a variant). Reproduced when the averaged reconstructed prior's Delta D_KL is negative for rare-trained and positive for frequent-trained models, and its magnitude is below the neural across-mouse mean magnitude of fig6d-prior-reconstruction in both conditions.
- edf3-simulated-posterior-population
A simulated population that stochastically encodes samples from the rare or frequent morph posterior reproduces the recorded remapping patterns.
Simulating neural populations that were designed to stochastically encode the posterior distributions also reproduced the remapping patterns described above
Results, CA1 population activity approximates optimal estimation of the stimulus, paragraph 6; Extended Data Fig. 3How it will be judged
Methods, Simulating a neural posterior distribution. Per trial, samples s from the group posterior (M8) at the trial's f-hat_h; neuron i has gamma activity with theta = 1 and kappa_i(s) = alpha exp(-(s - mu_i)^2/sigma^2) + beta, sigma = 4.11, alpha = 2, beta = 0, mu_i tiling f_h; 1,000 simulations (Extended Data Fig. 3g,h). The trials are those of the recorded sessions 8 to N (code TheoreticalSimMats uses a session's morph values and its cell count; the code's rbf_sigma 0.4, alpha 1 and beta 0.25 are a variant). Open, the samples per trial (Methods "several times", code 1) and the number of neurons, read as (a) one sample, the session's cell count, (b) 45 samples (one per position bin), 100 neurons. Reproduced when the z-scored rare minus frequent simulated similarity difference (Extended Data Fig. 3j) has the same sign as the recorded one (fig3f-zscored-difference-at-025) at every entry of the S-tilde = 0.25 row that was significant there, under both readings.
- supp-earlier-morph-studies
With reasonable assumed priors, the probabilistic inference framework reproduces the discrete and gradual remapping reported in earlier morphed-environment studies (Wills 2005, Colgin 2010 double and single location, Leutgeb 2005).
making reasonable assumptions, we reproduced each of the conflicting findings from previous studies using our probabilistic inference framework
Discussion, paragraph 1; Supplementary Figs. 1 to 4How it will be judged
A simulation with priors the authors propose for other laboratories' experiments (Supplementary Figs. 1 to 4; code WillsLeutgebColgin_models). It uses no data of this paper, so the dataset cannot test it; it is listed to cover the paper's findings.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none