Fig. 2

Validation: leave-one-batch-out projections. Data from a series of tests projecting a query consisting of cells from one batch, onto an atlas computed from cells from all other batches. For assessing robustness, we use the labels for each cells as inferred from analysis of the atlas including all batches as shown in Fig. 1 above. A Counting cells classified according to their reference annotated type and the type assigned to their query metacell by MCProj. Summary of false-positive and false-negative associations are shown as bar-plots (left—fraction of misclassified query cells per MCProj assigned type, bottom—fraction of misclassified query cells per original type). B The fraction of query cells which were not successfully projected and which were assigned a type different than the expected type, for each of the methods we compared. C Example of query metacells observed vs. projected expression over all genes. D For some example genes, shown are observed vs. projected expression levels over all query metacells (overlaying data from all batches)