Fig. 4: Stimulus- vs. Model-based temporal response functions.
A Analysis pipeline. In the classic stimulus-based temporal response function approach (top), a linear filter estimates the amount of variance explained by stimuli in the MEG activity. In our model-based temporal response function approach (bottom), the filter is used to estimate the amount of variance explained by the model dynamics. This approach requires a time-resolved, biologically plausible, model, which is the case of the Multisensory Correlation Detector. The method estimates the amount of MEG responses variance explained by each component of the Multisensory Correlation Detector. For each component, a temporal response function is fitted by ridge regression such that the convolution between single trial component activity and the temporal response function maximizes the correlation between the predicted MEG signal and the true MEG signal. In order to avoid over-fitting, the ridge regression quality is evaluated through cross-validation. B The amount of MEG explained variance is resolved in model components, sensors, and tasks. The table reads as follows: in the causality judgment blocks (third row), the MCDA input component (first column) explains a significant proportion of variance in a bilateral central cluster, with a Pearson’s ⍴ = 0.16 (right). Overall, MCDA, MCDV and MCDCORR explain MEG variance consistently across causality and temporal order judgment blocks. To the contrary, MCDLAG significantly fit the data only in the temporal order judgment blocks. Statistical significance of the correlation between cross-validated predicted MEG and true MEG was assessed via corrected cluster permutations. Significant sensors are highlighted in white. Error bars represent 2 s.e.m. across participants (N=13). Dots represent individual participants. C Source estimations (inflated brain) locate MCDA in bilateral posterior superior temporal gyrus, MCDV in bilateral occipital cortices, MCDCORR in bilateral superior temporal gyrus, bilateral posterior superior temporal sulcus and bilateral supramarginal gyrus. MCDLAG is not associated with any robust phase-locked activation.
