Extended Data Fig. 1: Data quality control.

a, Box plots depicting the probability of recall for items presented at each serial position. Primacy and recency effects are visible, consistent with expectations for performance in the free recall episodic memory paradigm. Whiskers on box plots represent maximum and minimum values. Boxes extend from the 25th to the 75th percentiles, the center lines represent the median. Loess regression with confidence intervals is superimposed to depict the overall distribution. Smooth curves are shown with 95% confidence bands b, Lag conditional response probability curves in our data (lag CRP), indicating expected temporal clustering behavior. Loess regression with confidence intervals depicts the overall distribution. Smooth curves are shown with 95% confidence bands. c, Boxplot showing the comparison of within-subject variance (across all measured electrodes at each band, blue box plot,) with the variance across subjects (at each band, yellow box plot). Across subjects variance is significantly greater than within-subject variance. Reported p-value from Wilcoxon rank sum test (one-sided with alternative greater). Boxplots extend from the 25th to the 75th percentiles, the center lines represent the median. d, Scatter plot showing the fraction of all BA38 electrodes exhibiting a significant subsequent memory effect at each frequency. We observed significant differences predicting recall success across the frequency spectrum, including the delta and gamma bands. Loess regression with confidence intervals depicts the overall distribution. Smooth curves are shown with 95% confidence bands. e, Distribution of SME values for each brain oscillation and cross-correlation based on Spearman’s rank correlation. f, Barplots showing the fraction of electrodes at which oscillations were detected in each frequency band in the recalled and non-recalled conditions. 85% of electrodes exhibited an oscillation in at least one of the delta, theta, or alpha frequency bands. g, Scatter plot showing individual electrode examples of power curves used for oscillation detection via the MODAL algorithm, both before and after subtraction of the best fit line. h, Principal component analysis of the subjects used for the within-subject analysis. Variance explained by each principal component is highlighted in the axis. i, Barplot showing the variance explained by each covariate adjusted across 10 principal components (wVE) for the within-subject data. Technical, biological and sequencing covariates calculated by PICARD (see Methods) are included. l, Principal component analysis of all the subjects used in this study. PMep = post-mortem epileptic subjects, UT = within-subjects, PMctl = post-mortem healthy subjects. m, Variance explained by each covariate adjusted across 10 principal components (wVE). Type corresponds to the three different types of data included in the analysis (PMep, UT, PMctl). Technical, biological and sequencing covariates calculated by PICARD (see Methods) are included. n, Association between the first two components and covariates based on adjusted gene expression. X-axis corresponds to the -log10(P-value) from linear regression modeling between PCs and covariates.