Auditory oddball analyses

For the analysis of EEG during the auditory oddball paradigm, data were further low-pass filtered at 20 Hz and baseline corrected over the first 800 ms (from the beginning of the trial to the onset of the fifth sound).

ERP topographies: We first analyzed group-level event-related potentials (ERP) elicited by the auditory oddball paradigm. These were obtained by averaging trials of each condition with the trimmed mean 80% (across trial) over the scalp channels at the subject-level. We then computed group-level ERP to deviant minus standard trials, before and after tDCS, and compared R+ with R− (or VS/UWS with MCS & exit-MCS) with the following contrast: [[Deviant − Standard]_after − [Deviant − Standard]_before]R+ vs. [[Deviant − Standard]_after − [Deviant − Standard]_before]R−. Results are reported by the mean ± standard error of the mean.

Temporal generalization decoding: As classical ERP analyses are prone to a high inter-subject spatial variability, especially in brain-lesioned patients, we completed the topographical analysis with the analysis of the temporal dynamic of the ERP. To that end, with used the Temporal Generalization decoding method described by King et al.⁵³. This MVPA approach relies on training a classifier at each time point of the trial to distinguish deviant from standard trials at the single-subject level. Each classifier was then tested not only on the time sample it was trained on, but also on every other samples of the trial to see if decoding performances generalized in time. This procedure thus allows to extract patterns of brain activation associated with different cognitive tasks based on their temporal dynamics. This procedure has previously been applied to an auditory oddball paradigm and showed that the violation of auditory regularity led to two kinds of neural activation patterns: an automatic, early and short-lived response to auditory novelty and a late and sustained (200–700 ms) pattern associated with the conscious access to the auditory novelty. This latter pattern probably reflects the same process as the one engaged in the generation of the P300 during the same paradigm⁵⁶. Here we applied this MVPA procedure using linear support vector classifier (with 10 iterations of a stratified 5-fold cross-validation). Performances of the classifiers were expressed as the area under the curve (AUC).