Fig. 3: Benchmarking results on pancreatic islet dataset.

Created in BioRender. He, C. (2025) https://BioRender.com/f1hibjwa Bar plot showing the absolute cell count for each cell type and the distribution of batches across different cell types. For visual clarity, only scVI and Seurat are included in the visualization, as scDML was unable to identify rare immune cells. b Scatterplot of average bio-conservation score against the average batch correction score for each method. c UMAP visualization contrasting the unintegrated pancreatic islet dataset with its integrated counterparts using scCRAFT and two other most popular benchmarking methods. The top panel shows the integration of different batches (sequencing technologies), the middle panel shows the cell types, and the bottom panel shows the distribution of negative, positive, and true positive cells. Positive cells are those surrounded by cells of the same type, while true positive cells are a subset of positive cells whose surrounding batch distribution matches the global batch distribution for that cell type. d UMAP visualization focusing on the integration of rare cell types (T cells, macrophages, and mast cells) for scCRAFT, scVI, and Harmony. For each method, the left side of the panel shows the cell type distribution, and the right side shows the classification of cells into negative, positive, and true positive.