Fig. 5: OnClass can assign marker genes to seen and unseen cell types.
From: Leveraging the Cell Ontology to classify unseen cell types
a 2-D UMAP plot showing OnClass’s integration of 26 datasets35 on 6 groups. We combined the 9 cell types (list here) into six groups (Neurons, PBMCs, Pancreatic islets, HSCs, Jurkat+293T, and Macrophages). b Box plot showing the comparison between OnClass and expression on data integration in terms of the silhouette coefficient (P-value < 7e−293). P-value is determined using a two-sided t-test (n = 6 cell types). Minima, maxima, centre, bounds of box, and whiskers represent quantile 1–1.5*interquartile range (IQR), quantile 3+1.5*IQR, median, quantile 1, and quantile 3. c Bar plot showing the AUROC of predicting marker genes using different datasets. Error bar represents standard errors across 52 unseen cell types and 17 seen cell types for Muris FACS, 48 unseen cell types and 21 seen cell types for Muris droplet, 55 unseen cell types and 13 seen cell types for Lemur 1, 46 unseen cell types and 22 seen cell types for Lemur 2, 55 unseen cell types and 13 seen cell types for Lemur 3, and 46 unseen cell types and 22 seen cell types for Lemur 4. Mean is used to measure the centre for the error bar. d Bar plot comparing the AUROC of using OnClass-computed marker genes and curated marker genes to classify cells in different datasets using marker genes obtained from Lemur 2 (d) and Lemur 4 (e). Error bar represents standard errors of n = 17, 21, 13, 22, 13, 22 for Muris FACS, Muris droplet, Lemur 1, Lemur 2, Lemur 3, Lemur 4, respectively. Mean is used to measure the centre for the error bar. f Heatmap showing the AUROC of using marker genes to classify cells in the cross-dataset setting. The x-axis is the test set and the y-axis is the training set.
