PAC as markers of correct vs. incorrect memory retention identified with machine learning.

(A) SVM decoding accuracy (accuracy minus chance—chance level: 0%) for a 2-class decoding analysis of PAC strength and SEEG contacts as features (correct vs. incorrect memory retention in all memory conditions). The colored bars represent accuracy minus chance for each participant (sorted as a function of accuracy with a jet colormap). Orange shaded rectangle overlaps with participants showing decoding accuracy significantly above chance. Blue shaded rectangle overlaps with participants with decoding accuracy not significantly different from chance. Asterisk: significant, ns: nonsignificant. Source data can be found at https://osf.io/m7dta/ (B) ROC for each participant (same color code as in A). Black dashed line represents the chance level. Source data can be found at https://osf.io/m7dta/. (C) Normalized feature weights showing features (SEEG contacts) with the largest influence (z-score) for each participant with significant decoding accuracy. Source data can be found at https://osf.io/m7dta/. PAC, phase amplitude coupling; ROC, receiver operating characteristic curve; SVM, support vector machine.