Figure 1.
Workflow of ProjectSVR. (a) The ProjectSVR takes integrated embeddings (e.g. UMAP) and gene set scores calculated via the UCell algorithm as inputs. (b) Then a regression model is fitted by supported vector regression (SVR). (c) For mapping, the query count matrix is transformed into signature scores, and the trained models predict query embeddings. Final predictions are aggregated via median. Cell type labels are transferred using a k-nearest neighbors (KNN) classifier.
