Fig. 1: OnClass overview.
From: Leveraging the Cell Ontology to classify unseen cell types
OnClass first maps user terminology to Cell Ontology terms based on the text embedding similarity. It then embeds cell ontology terms into a low-dimensional space using the Cell Ontology graph. Single-cell transcriptomes are then projected into the same low-dimensional space using a nonlinear transformation. Finally, OnClass refines the annotation of each cell by first overlaying confidence scores on the Cell Ontology graph and then propagating these scores using random walk with restart.
