Bayesian decoding of marmoset position from individual recording days.
(A) Plots of the number of recordings as a function of the fraction of trajectories per recording whose decoding on average explained a significant amount of variance of the true trajectory. Significance in this case was determined by p-value against the null distribution of recomputed decoded trajectories with shuffled place-cell activity. (B) Plots of the same analysis as in panel A but with the correlation of the true trajectory and decoded trajectory as the performance metric of choice. (C) Plots of the percent of variance explained by the decoded trajectories as a function of the number of neurons per recording session. Error bars are SEM computed across all recording days with the given number of neurons. (D) Plots of the same analysis as in panel C with the correlation of the true trajectory and decoded trajectory as the performance metric of choice. (E) The relationship between the mean explained variance and the mean correlation across all trajectories is shown for each recording session. Here, error bars in each dimension are SEM of explained variance and correlation computed across all trajectories for each recording session. (F) The mean absolute error between the decoded and true trajectories is shown for each recording session with a statistical significance threshold given at 0.05 by the vertical dashed line. Here, again, p-values were determined by shuffling of place-cell activity followed by recalculation of the mean absolute error. Significant decoding with respect to mean absolute error was found in all but 3 recording sessions (i.e., 34/37 sessions). (G) Plots of the mean absolute error for each recording session with true place fields on the left and the average of the null distribution for each recording on the right. The distributions, here, significantly differ as determined by Mann–Whitney U test (p = 1.3 × 10−5). This significance is indicated by the horizontal black bar and asterisk plotted above the distributions. The underlying data can be found at https://doi.org/10.5061/dryad.kk63d49.
