Fig. 1.

Steps in MVPA analysis. Note. A) Participants’ time-series data were converted to a vector (one vector per run). Events were labelled by stimulus emotion (hypothetical events represented as coloured bars; orange = anger, yellow = fear, light green = happiness, light blue = sadness, dark grey = neutral). The hemodynamic response function was fit to each labelled time-series. The first-order Legendre polynomial was removed and data were Z-scored in this step. B) Data were divided into separate vectors for each emotion and stacked by participant. Participants’ stacked vectors were labelled with their epilepsy status (purple = typically developing [TD] youth; green = youth with focal epilepsy [FE]. C) Using a leave-one-out cross-validation approach, emotion-specific vectors were recursively split into training and test sets (where a test set consisted of all runs for a single participant). Using a roaming searchlight, a support vector machine was trained at each voxel. Classifier performance was calculated at each voxel as the accuracy with which it labelled each test set. This procedure resulted in an average accuracy brain map for each emotion. The process in C was repeated for each emotion category. D) Accuracy maps for each emotion were averaged to create a grand average map. This brain map represented the classifier’s performance across all emotions in the task. Please see 2.4.2 for additional details. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)