Data and code for: Spatial cell type enrichment predicts mouse brain connectivity

A fundamental neuroscience topic is the link between the brain's
molecular, cellular and cytoarchitectonic properties and structural
connectivity (SC). Recent studies relate inter-regional connectivity to
gene expression, but the relationship to regional cell-type distributions
remains understudied. Here, we utilize whole-brain mapping of neuronal and
non-neuronal subtypes via the Matrix Inversion and Subset Selection (MISS)
algorithm to model inter-regional connectivity as a function of regional
cell-type composition with machine learning. We deployed random forest
algorithms for predicting connectivity from cell type densities,
demonstrating surprisingly strong prediction accuracy of cell types in
general and particular cells like oligodendrocytes. We found evidence of a
strong distance-dependency in the cell-connectivity relationship, with
layer-specific excitatory neurons contributing the most for long-range
connectivity, while vascular and astroglia are salient for short-range
connections. Our results demonstrate a link between cell types and
connectivity, providing a roadmap for examining this relationship in other
species, including humans.