Figure 1From: Greedy low-rank algorithm for spatial connectome regressionIn this paper, we present an improved method for the mesoscale connectome inference problem. (A) The goal is to find a voxel-by-voxel matrix W so that the pattern of neural projections y arising from an injection x is reproduced by matrix-vector multiplication, \(y \approx W x\). The vectors x and y contain the fraction of fluorescing pixels in each voxel from viral tracing experiments. (B) An example of the data, in this case a coronal slice from a tracing experiment delivered to primary motor cortex (MOp). Bright green areas are neural cells expressing the green fluorescent protein. (C) The raw data are preprocessed to separate the injection site (red/orange) from its projections (green). Fluorescence values in the injection site enter into the source vector x, whereas fluorescence everywhere else is stored in the entries of the target vector y. The x and y are discretized volume images of the brain reshaped into vector form. Entry \(W_{ij}\) models the expected fluorescence at location i from one unit of source fluorescence at location j, a linear operator mapping from source images to target images. Image credit (B and C): Allen Institute for Brain ScienceBack to article page