Figure 3. Task-specific predictive contributions of IPL subregions.

Pattern-learning algorithms extracted predictive rules from neural activity estimates aggregated in the left vs. right anterior vs. posterior IPL subregions from the three target experimental conditions for attentional reorienting (top), lexical decisions (center), and perspective taking of others’ mental states (bottom). Colors show the predictive signature with relative contributions of each of the four IPL subregions in detecting the presence of the three cognitive states from neural activity responses. A more positive subregion weight (brown color) for a given task implies that neural activity from this subregion carried information that increased the probability of a specific task being represented in trial brain scans. A leave-one-subject-out cross-validation was implemented to fit the predictive model. Negative values (cold colors) denote driving the prediction decision toward the respective other tasks.

Figure 3—figure supplement 1. Neural activity estimates for the target and control conditions of the three tasks.

For visualization purposes, beta estimates were extracted from GLM_cond at the center of mass for each IPL subregion.