Fig. 2: Benchmarking results on simulation and real data.

Created in BioRender. He, C. (2025) https://BioRender.com/6tagtjla UMAP visualization contrasting unintegrated simulation datasets with scCRAFT-integrated counterparts under various simulation tasks, including baseline where cell types are shared equally across batches of the same size, unbalanced batch sizes (batch-0.125), batch-specific cell types (common-3), and rare cell types (rare-0.1). The upper part of each box shows the integration of different batches, and the lower part shows the delineation of cell types (Groups) within these integrated datasets. Overview of the quantitative evaluation of 10 simulations tasks (b) and 9 real datasets (c), including the specific rankings and scores for each evaluation metric across 7 benchmarking methods. The average scores across individual metrics for each dataset are illustrated with orange and blue circles. The average scores across all datasets are illustrated with orange and blue bars. The composite scores for each dataset are illustrated with green circles, and were calculated as the weighted average of bio-conservation and batch mixing, with a 60/40 ratio favoring the former over the latter. The overall score represents the average of the composite scores of all datasets. The diameter of each circle represents a linear rescaling of its corresponding values to a range between 0.1 and 1. The length of each bar represents the absolute average value. Methods are ranked based on their overall score, from highest to lowest.