Fig. 4: Anatomical characterization of fully reconstructed neurons.

A Heatmap of pairwise neuron similarities. Each row and column are individual neuron, with color showing similarity values calculated as the product of the cosine distance between standardized morphological features over the exponential of normalized between-soma distance. Neurons are categorized into four clusters (C1, C2, C3, and C4) using Spectral clustering (see Methods). B Horizontal projections on the CCFv3 template of five randomly selected neurons. C A pair-plot displaying the composition of neuron types within each cluster (pie plots in the main diagonal). Soma spatial distributions of cluster pairs are shown in the upper triangle, while 3D scatter plots (lower triangle) show pairwise separability of neurons from each cluster (color-coded) with respect to the top 3 discriminating features between cluster pairs. The average Silhouette Coefficients (SC) are specified in red. Viewpoints of the scatter plots are optimized for cluster separation. D Heatmap of the number of times (hit rate) a feature was selected by mRMR as a top three discriminating feature of the clusters in six independent rounds. Each round corresponded to a separate cluster pair. E Top, box plot of the top-ranking feature (“tilt_remote_std”) of neurons between clusters. Bottom, density plot of maximal Euclidean bifurcation-to-soma distance across neurons in each cluster. The neuron numbers for C1, C2, C3, and C4 are 502, 515, 499, and 360. F Matrix visualization of the mean (light green) and standard deviation (std; light blue) of branch numbers (represented as dot size). Each row corresponds to one cluster, and each column represents the distance interval (300 µm) at which we measured branch numbers. Source data are provided as a Source Data file. Box plot: edges, 25th and 75th percentiles; central line, the median; whiskers, 1.5×the interquartile range of the edges; dots, outliers.