used to evaluate a given cell-type, the method generates a network based on the Spearman correlation between all cells across the genes within the set. The correlation is rank standardized to provide network weightings between each pair of cells, and then a neighbor voting predictor scores cells as possessing a given annotation. The score is calculated as the sum of a given cell’s connectivity weighting to neighbors possessing a given cell annotation. For cross-validation, we permute through all possible combinations of leave-one-batch-out cross-validation, and report the degree to which cells of the same type are recovered as the mean area under the receiver operator characteristic curve (AUROC) across all folds. To improve speed, AUROCs are calculated analytically: AUCj=(∑(i∣Celli∈∣Celltypej)Ranksi-NPos∗(NPos+1)2NPos∗NNeg) where “Ranks” are the ranks of the hidden positives, Npos is the number of true positives, and NNeg is the number of true negatives.
