Fig. 2: Cell morphology spaces accurately summarize the complexity of cell shapes.

a UMAP representation of the cell morphology space of 506 neurons from the murine visual cortex profiled with whole-cell patch-clamp. The representation is colored by the morphological cell populations that resulted from clustering the cell morphology space using Louvain community detection. The morphologies of individual neurons randomly sampled from 4 of the populations are shown for reference. Apical and basal dendrites are indicated in purple and red, respectively. b The UMAP representation is colored by the neuronal type (top), cortical layer (middle), and Cre driver line (bottom). The GW cell morphology space captures morphological differences between neurons of different molecular type and anatomic location. c The metric structure of the GW morphology space enables performing algebraic operations such as averaging shapes. The figure shows the medoid (indicated with a circle) and average morphologies (in boxes) computed for each of the morphological populations in (a). d Cell-type separation (CTS) score, accuracy, and Matthews correlation coefficient (MCC) for CAJAL and six other neuronal morphometry methods in the task of predicting the molecular and/or anatomic location of individual neurons in four datasets (one patch-clamp, two Patch-seq, and one fMOST datasets). The dashed line indicates the accuracy of a random classifier. More detailed information is presented in Supplementary Tables 1 and 2. Source data are provided as a Source data file.