Fig. 3

Benchmark results when classifying distinct cell populations in six pancreas datasets. The benchmark was performed in a sixfold cross-validation scheme where one dataset was used for training and the other five for testing. All metrics were ordered by the tools’ mean accuracy. (*) indicates results where ambiguous classifications were accepted as correct. Garnett-markers and Garnett-DEG are two evaluations of Garnett, one using the same markers as scAnnotatR (Garnett-markers) and one using top 10 DEG (Garnett-DEG). scibet-rej evaluates SciBet’s performance using a confidence threshold of 0.4. a The prediction accuracy in each training and test set for all tools. b Average sensitivity and c specificity as the mean sensitivity and specificity, respectively, across all classified cell types for each dataset and iteration. d Proportion of cells not present in the training data, that were correctly not-classified