Fig. 5: Comparison of QDyeFinder versus ground truth with manual tracing.
a L2/3 neurons in S1 were labelled with 7-colour Tetbow via in utero electroporation (age, P28) (left). 35 neurons were then manually traced via Neurolucida 360. They were then split into fragments (no branches). Each fragment is represented by a different colour (right). Somata are also highlighted, but were excluded from the analyses. Image taken with a ×63 objective lens (NA = 1.30). Z-stacked images of 511.36 × 512.99 × 47.99 μm3 are shown. b A representative example neuron showing the relevant fragments (left), the vector normalised colour vector of each trace (middle), and the Euclidean distance to the mean for the neuron (right). Red bars indicates mean ± SD and blue box plots indicate median ± interquartile range (IQR) of d to the centroid of the colour vectors for all fragments (n = 47 fragments). See Supplementary Data 1 for all the results. c The ground truth of each neuron and their neurite fragments in physical space (left), and in UMAP-reduced colour space (right). d The optimum threshold was calculated by running the dCrawler at a Th(d) ranging from 0.05 to 1, and calculating F1 score for each neuron at each Th(d) (gray dots). The median F1 score for each Th(d) is also displayed (red line). The optimum Th(d) was calculated to be 0.2. e The dCrawler clustering of the neurite fragments at the the optimum Th(d), shown in physical space (left), and in UMAP-reduced colour space (right panel). f Representative neurons (ground truth in blue lines, top row) paired to their best cluster. Fragments in both the ground truth and the dCrawler cluster (bottom row) are considered correct (green), those only in the dCrawler cluster are a false positive (red), those in the ground truth only as a false negative (black), and those in neither the ground truth or dCrawler cluster as a true negative (grey). See Supplementary Data 2 for all the results. Source data are provided as a Source Data file.
