Fig. 4.

Temporal patterns of changes in seizure pathways. (A) Calculation of patient 931’s temporal correlation pattern. Temporal distances between seizures, which were derived from the patient’s seizure times (Top), were compared to seizure dissimilarities at different timescales (Middle). Example timescales T = 1 d, T = 3 d, and T = 5 d are shown (scatterplots; Middle). In each scatterplot, brown shading indicates the timescale, black points correspond to seizure pairs used to compute the correlation for that timescale, and gray points were pairs excluded from the correlation computation. At T = 5 d, all seizure pairs are included, producing the same temporal correlation as in Fig. 3B. Scanning the timescale produces a set of correlations, or temporal correlation pattern, shown in the heat map (Bottom). Gray dots in the heat map indicate insufficient information at that timescale, and these timescales are excluded from downstream analysis. (B) Seizure dissimilarities were modeled based on linear (Left), circadian (Middle), or a combination of linear + circadian (Right) changes in seizure pathways. The simulated changes in seizure pathways are shown as different functions of time (Top), with patient 931’s seizures marked in orange. Temporal distances and simulated seizure dissimilarities were compared across different timescales (scatterplots; Middle), yielding a temporal correlation pattern for each model (heat maps; Bottom). Example models and simulation results are shown here; the full set of tested models is provided in SI Appendix, Table S10. (C) Observed temporal correlation patterns of seizure pathways in each patient, categorized by the model that best reproduces these dynamics. The goodness of model fit was measured using model likelihood (gray heat map). The full details of each patient’s model are provided in SI Appendix, Fig. S10.2.