Extended Data Fig. 7. Performance comparison using batch integration evaluation metrics.

Median performance of SATURN and baseline methods evaluated using (a) weighted mean of the batch removal score (Avg Batch), (b) bio-conservation score (Avg Bio), and (c) overall score. Distribution is estimated with n = 30 runs of each method. Overall score is calculated as (0.6 * Avg Bio) + (0.4 * Avg Batch). Avg Bio is calculated as the average of NMI cell type score, ARI cell type score, and ASW cell type scores. Avg Batch is calculated as the average of ASW batch score, and graph connectivity scores, where species is taken as the batch variable. Neighbor calculation is done using default Scanpy settings, using each methods’ embeddings. Score calculation is done using default SCIB settings³².