Fig. 2: Intent model. | Nature Communications

Fig. 2: Intent model.

From: Neural signatures of emotional intent and inference align during social consensus

Fig. 2

a Training. The model was trained to predict the target’s self-reported emotional intensity from the observer’s brain. A prediction-outcome correlation was calculated across each cross-validated (CV) training fold (average Pearson’s r = 0.65 ± 0.34 STD across folds is significant in a two-tailed one-sample t-test: t(99) = 18.91, p < 0.001, CI = [0.58 – 0.71]). The target’s true self-reported intensity level (y) is plotted on the x-axis. The model’s prediction of the intensity level (y-fit) is plotted on the y-axis. b Validation. The model was validated by calculating the prediction-outcome correlation across five levels of self-reported intensity within participants in held-out validation trials (see Fig. S5). Average r values for each participant (N = 100) are plotted on the y-axis (orange bar). The model is sensitive to its validation set (t(99) = 9.65, P < 0.001) in a two-tailed one-sample t-test. To test its specificity, we repeated this analysis in the inference validation set (gray bar). The intent model had a higher fit on the intent validation set than the inference set, but the difference was not significant (see Fig. S10A for internal validations). Data are represented in box plots where the median is a black line and the upper and lower whiskers represent the bounds of the quartiles. c Feature importance. The model’s voxel-weight map (Fig. 1b) is loosely thresholded (P < 0.01 uncorrected (unc.)) based on a bootstrap (5000 samples) hypothesis test and plotted for visualization (FDR thresholded regions in Table S1). Activity in right visual and right anterior insular cortices, as well as right angular gyrus, left posterior cingulate cortex (PCC), bilateral precuneus, and bilateral superior frontal gyrus were most important for this prediction. d Associated NeuroSynth terms. The unthresholded predictive weight map was fed to the NeuroSynth topic map decoder. A word cloud was constructed of the top 20 terms (excluding singular brain regions). Word size is scaled by strength of similarity. The top five terms are colored orange.

Back to article page