Fig. 4: Identifying correlates between neural data and internal states.

a Subjects were implanted with multiple intracranial depth electrodes. The electrode coverage is visualized by mapping subject-specific coordinates of each channel onto a template brain called cvs_avg35_inMNI152 from Freesurfer⁹⁴. Electrodes are made up of multiple channels. For example, the channel clinically called X4 for subject 6, located in the right intraparietal sulcus (IPS R), is outlined in black where “A" means anterior and “P" means posterior. b The spectrogram of channel X4 in IPS R for subject 6 for three trials during Speed Instruction, time-locked to its onset (white solid line). It is represented by the time—frequency domain. The color of each pixel represents the normalized power. c Time-series of the error state (blue solid line) for subject 6. The black upward-pointing triangle marks the value of the error state for the trials corresponding to the spectrograms in (b). The non-parametric cluster statistic identified a significant cluster (outlined by the solid black line in (b)) where power correlated with the error state across all channels in the IPS R in the population. For example, the power in the cluster increases for large positive magnitudes of the error state (too slow) in trials 29 and 109 but decreases when the error state is large and negative (too fast) in trial 77. d The statistical map over the time-frequency domain from the non-parametric cluster statistic between the IPS R and the error state across the population during Speed Instruction. Each pixel represents the hierarchical average across channels and subjects of the two-tailed t-statistic of the two-tailed Spearman’s correlation between the error state and spectrogram. The white solid line markers where data was time-locked to epoch onset. Clusters were found as adjacent time-frequency windows that show a significant correlation (outlined in black). e There is a significant relationship (two-tailed Spearman’s correlation: r = − 0.31, p = 3.87 × 10⁻⁴) between the average normalized power in the cluster and the error state for the example in (b). This correlation coefficient also represents the encoding strength of this channel. The black star markers correspond to the trials in (c). The least-squares line is marked as the gray dashed line. Source data are provided as a Source Data file.