Fig. 2: RSA-based clustering identifies separate prefrontal and limbic food networks.
We applied a network clustering approach based on Representational Similarity Analysis (RSA15) to examine the representation of food images within food-responsive regions of the brain (A), (see Fig. 1). A We extracted and compared the multivariate food representations of each food from our food-responsive regions-of-interest (ROIs) to generate separate neural Representational Dissimilarity Matrices, B which we then compared across each pair of ROIs, resulting in a second-level ROI similarity matrix reflecting the similarity of representational profiles across our food-responsive regions. C, D We applied a network clustering algorithm to this matrix, which identified an optimal number of 2 clusters and partitioned the ROIs into two networks exhibiting categorically different representational profiles: a Prefrontal network composed of dorsal brain regions in prefrontal and parietal cortices, and a Limbic network composed of ventral cortico-limbic and sub-cortical brain regions. NB: Node sizes in this graph (D) are proportional to the average similarity of each node to every other node. The coloring of edge lines between nodes indicates whether those edges are within or between networks, and their thickness indicates the edge strength (i.e., similarity). The position of nodes within the graph indicates relative centrality within the whole node network. For copyright reasons, original images have been replaced with visually similar images from the public domain. Photos from Wikimedia.com. See the “Data availability” section for the location of original images.
