An algorithm for identifying task-specific brain subnetworks using the visuomotor system as an example
We describe an algorithm that identifies a subnetwork of brain regions involved in producing a task-specific behavior, here visuomotor behavior, from an anatomically defined primate brain connectome. The algorithm first finds the brain regions connected to an output region (here, primary motor cortex, M1) by one connection. It then identifies all regions, termed layer 2 regions, connected to these layer 1 regions by one connection. This process continues until the layer containing the input region (here, primary visual cortex, V1) is reached. The algorithm then finds, subject to a user-set maximum step number, all paths linking the input and output regions. The brain regions in these paths constitute the initial subnetwork identification that performs the task. Regions known not to be task-involved (for example, regions in the ventral stream of visual information vs. the dorsal stream, which helps generate visuomotor behavior) are then removed. Structural subnetwork analysis showed that the intraparietal sulcus of the parietal cortex (PCIP) was most, and the secondary visual (V2) and superior parietal (PCS) cortices second-most, central to local network activity. Changing PCIP, V2 and PCS activity was thus most likely to alter activity of the entire subnetwork. Model sufficiency was tested by instantiating each brain region inherent activity with multiple versions of a simple two-dimensional (2D) model that can produce oscillatory activity and synaptically interconnecting the regions to produce a macroscopic visuomotor model. The model reproduced the experimental local field potential (LFP) activity of the brain regions identified as part of the visuomotor subnetwork.