Evaluating hippocampal replay without a ground truth

  1. Masahiro Takigawa  Is a corresponding author
  2. Marta Huelin Gorriz
  3. Margot Tirole
  4. Daniel Bendor  Is a corresponding author
  1. Institute of Behavioural Neuroscience (IBN), University College London (UCL), United Kingdom
11 figures and 1 additional file

Figures

Figure 1 with 4 supplements
Demonstration of novel replay analysis framework for comparing sequence fidelity with track discriminability.

(A) Experimental design. For each recording session, the animal ran back and forth on two novel linear tracks (RUN) with resting sessions before (PRE) and afterward (POST). Rat schematic in A was …

Figure 1—source data 1

Summary of replay detection performance for weighted correlation method with two shuffles.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig1-data1-v2.xlsx
Figure 1—figure supplement 1
Individual examples of decoded trajectories with different sequence fidelity and reactivation bias.

Each panel is the posterior probabilities of a candidate event decoded against track 1 template (left) or track 2 template (right), with the y axis representing position and the x axis representing …

Figure 1—figure supplement 2
Mean log odds difference and proportion of significant events detected hold across 10 sessions.

(A,B) The proportion of significant events and mean log odds difference at (A) an alpha level of 0.05 and (B) an FPR-matched alpha level with a mean false-positive rate of 5%.

The shaded box indicated 95% bootstrap confidence interval. Different colors are used to indicate different behavioral states: RUN (blue) and POST (orange). Different symbols are used to indicate different sessions. (Number of candidate replay events for sessions 1-10 : RUN n = 579, 559, 500, 388, 483, 400, 466, 459, 388 and 421 and POST n = 778, 1228, 969, 1770, 1578, 1986, 1793, 1687, 1278 and 1259, respectively)

Figure 1—figure supplement 3
Schematics of shuffling procedures performed to obtain four null distributions for replay detection.

(A) A spike train circular shuffle (dark blue) in which each cell’s spike train was independently circularly shifted in time by a random amount within each replay event. (B) A place field circular …

Figure 1—figure supplement 4
Proportion of false-positive events detected when using different shuffle methods and randomized datasets for a null distribution.

The detection during RUN and POST was based on single shuffle - (A) time bin permutation shuffle, (B) spike train circular shift shuffle, (C) place field circular shift shuffle, and (D) place bin …

Figure 2 with 4 supplements
Replay detection performance improves with ripple power.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) as ripple power increases (0–3, 3–5, 5–10, 10, and above). The shaded region indicates …

Figure 2—source data 1

Summary of replay detection performance at different ripple power thresholds.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig2-data1-v2.xlsx
Figure 2—figure supplement 1
The replay event distribution, number of active place cells, and total spikes during replay event at different ripple powers.

(A) The distribution of candidate replay events across different ripple powers. (B) The proportion of significant events (out of all candidate events) at p-value≤0.05 at different ripple powers. (C) …

Figure 2—figure supplement 2
The mean false-positive rate across both tracks for replay events detected with different ripple power range.

(A,B) The mean false-positive rate calculated at different alpha levels (i.e. 0.05, 0.02, 0.01, 0.005, 0.002, 0.001) as ripple power increased (i.e. 0–3, 3–5, 5–10,10, and above). The error bar …

Figure 2—figure supplement 3
Replay detection performance improves with ripple power using stricter criteria for candidate events.

The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) as ripple power increases (0–3, 3–5, 5–10, 10, and above). (A,B) A stricter replay event …

Figure 2—figure supplement 4
Replay detection performance improves with ripple threshold for POST but not RUN.

(A,B) The proportion of significant events (out of all candidate events based on multi-unit activity [MUA] criteria alone) and mean log odds difference at different alpha level (0.2–0.001) as ripple …

Figure 3 with 1 supplement
Replay detection performance was sensitive to the shuffling method applied.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when using four different shuffling methods: (1) spike train circular shuffle (dark …

Figure 3—source data 1

Summary of replay detection performance using four different shuflling methods.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig3-data1-v2.xlsx
Figure 3—figure supplement 1
The mean false-positive rate across both tracks for replay events detected using different shuffling methods.

(A) A spike train circular shuffle (dark blue) in which each cell’s spike train was independently circularly shifted in time by a random amount within each replay event. (B) A place field circular …

Figure 4 with 2 supplements
Replay detection performance can be improved by adding stricter detection criteria.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) with four different detection criteria: (1) Only a post-decoding place bin circular …

Figure 4—source data 1

Summary of replay detection performance when adding stricter detection criteria.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig4-data1-v2.xlsx
Figure 4—figure supplement 1
The mean false-positive rate across both tracks for replay events detected using different detection criteria.

(A,B) The mean false-positive rate calculated at different alpha levels (i.e. 0.05, 0.02, 0.01, 0.005, 0.002, 0.001) using four different detection criteria: (1) Only a place bin circular shuffle, …

Figure 4—figure supplement 2
Comparison of the replay detection performance when applying different jump distance thresholds.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) using place bin circular shuffle plus jump distance threshold ranging from 0.2 to 1 …

Figure 5 with 1 supplement
The performance of rank-order-based replay detection method depends on the selection of spikes for analysis.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when (1) all spikes or (2) only the median spike fired by each place cell was included …

Figure 5—source data 1

Summary of replay detection performance for rank-order-based methods.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig5-data1-v2.xlsx
Figure 5—figure supplement 1
The mean false-positive rate across both tracks when all spikes or median spike fired by each place cell was included for Spearman’s rank-order-based analysis.

(A,B) The mean false-positive rate calculated at different alpha levels (i.e. 0.05, 0.02, 0.01, 0.005, 0.002, 0.001) when (1) all spikes or (2) only the median spike fired by each place cell was …

Figure 6 with 6 supplements
Comparison of different replay detection methods for replay events during PRE, RUN, and POST.

(A,B) The proportion of significant events and mean log odds difference at (A) an alpha level = 0.05 and (B) an FPR-matched alpha level with a mean false-positive rate of 5% using a range of …

Figure 6—source data 1

Summary of replay detection performance across multiple detection methods.

https://cdn.elifesciences.org/articles/85635/elife-85635-fig6-data1-v2.xlsx
Figure 6—figure supplement 1
Comparison of different replay detection methods for replay events during PRE, RUN, and POST when log odds difference was not shuffle-subtracted.

(A,B) The proportion of significant events and mean log odds difference at (A) an alpha level = 0.05 and (B) an FPR-matched alpha level with a mean false-positive rate of 5% using a range of …

Figure 6—figure supplement 2
Comparison of mean log odds difference using weighted correlation and linear fitting when the posterior probabilities were normalized within track or cross-track.

(A,B) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when using weighted correlation and linear fitting where the posterior probabilities …

Figure 6—figure supplement 3
Comparison of replay detection performance using weighted correlation for replay events during PRE, RUN, and POST.

(A–J) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when using weighted correlation with different detection criteria. (A–C) Only …

Figure 6—figure supplement 4
Comparison of replay detection performance using Spearman’s rank-order-based correlation for replay events during PRE, RUN, and POST.

(A–F) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when (A–C) all spikes or (D–F) only the median spike fired by each place cell was …

Figure 6—figure supplement 5
Comparison of replay detection performance using linear fitting for replay events during PRE, RUN, and POST.

(A–F) The proportion of significant events and mean log odds difference at different alpha levels (0.2–0.001) when using linear fitting approach with (A–C) only post-decoding place bin circular …

Figure 6—figure supplement 6
Comparison of the proportion of multi-track events and mean false-positive rate for different replay detection methods, and for replay events during PRE, RUN, and POST.

Multi-track events are replay events that pass all shuffling tests and/or additional replay criteria used for both tracks. At an alpha level = 0.05, shuffling methods tested include: (1) weighted …

Author response image 1
Distribution of Spearman’s rank order correlation score and p value for false events with random sequence where each neuron fires one (left), two (middle) or three (right) spikes.
Author response image 2
Distribution of Spearman’s rank order correlation score and p value for mixture of 20% true events and 80% false events where each neuron fires one (left), two (middle) or three (right) spikes.
Author response image 3
Number of true events (blue) and false events (yellow) detected based on α level 0.

05 (upper left), empirical false positive rate 5% (upper right) and false discovery rate 5% (lower left, based on BH method).

Author response image 4
Proportion of false events detected when using dataset with within and cross experiment cell-id randomization and place field randomization.

The detection was based on single shuffle including time bin permutation shuffle, spike train circular shift shuffle, place field circular shift shuffle, and place bin circular shift shuffle.

Author response image 5
Mean log odds and proportion of significant events at α level 0.

05 or FDR-matched α level for individual sessions when replay events were detected using (A-B) weighted correlation with two shuffles and (C-D) Spearman correlation using only each cell’s median …

Additional files

Download links