Figure 3: Classification of projection existence from neuron density similarity and Euclidean distance.

(A) Posterior probability of a projection being present resulting from training the classifier on all projections. Black lines are positioned at p(present) = 0.85 and p(present) = 0.15. Also see Supplementary Table S1 for predictions made about unsampled projections at these thresholds. (B) Cross-validated classification performance at different thresholds. Mean prediction accuracy for projections that were predicted to be present and absent (light green) as well as overall mean prediction accuracy (dark green) are shown. Also shown is the fraction of the test set that was classified at each threshold (black). Error bars indicate standard deviations. (C) Youden-index J for all combinations of parameters. Distribution of mean J across thresholds p(present) = 0.85 to p(present) = 1.00 for all 100 rounds of cross-validation. Boxplots indicate median J by a black bar and outliers by gray circles.