Fig. 4: TDANNs replicates animacy but not action-related organization of OTC.

a Spatial distribution of each category (as defined by t-values) on the simulated cortical space of the VTC-like layer of five random initializations of the TDANN. Rows correspond to each of the five initializations. Stars represent the location of the top-50 most selective units for that category. Category-selective units (positive t values) are shown in red, while units not selective for that category (negative t values) are shown in blue. b Overlap analysis. Statistical significance was assessed using permutation tests (10,000 randomizations on the mean overlap score across initializations). Stars represent statistical significance at the minimum resolvable p-value (p = 0.0001), corresponding to the 10,000 permutation limit. Error bars correspond to ± 1 SEM across the random initializations. Black dashed line represents baseline (overlap of 0.5 means no correlation between the presence of two categories). Each data point represents the value from a single TDANN initialization (n = 5 model initializations). c Selectivity profile of the top-50 most selective units for each category (red = faces; orange = bodies; yellow = hands; dark blue = tools; blue = manipulable objects; light blue = non-manipulable objects), based on the activation of the VTC-like layer (as in a). Each data point corresponds to one TDANN model initialization (n = 5 model initializations). Error bars indicate ± 1 SEM across model initializations. A baseline overlap of 0.5 denotes chance-level correspondence between category-selective units. Source data are provided as a Source Data file.