Fig. 4: Training with different datasets and proportions of unseen cell types highlights OnClass versatility and accuracy.
From: Leveraging the Cell Ontology to classify unseen cell types
a–d Heatmaps showing the ratio of unseen cell types (a), Accuracy@3 (b), unseen AUROC (c), and unseen AUPRC (d) in cross-dataset prediction. The x-axis is the test set and the y-axis is the training set. e Bar plot showing the AUROC of OnClass on 9 cell types, including 4 present in the training set (green) and 5 not present in the training set (yellow). f Bar plot showing the AUROC of OnClass using all six datasets (red) and each individual dataset (blue) as the training set to classify cells in the 26-dataset. Error bar represents standard errors. Mean is used to measure the centre for the error bar.
