Fig. 1

Schematic illustration of cluster similarity spectrum data representation. a First, clustering is performed on each sample (A, B, C) separately. Second, the average expression profile of each cluster is calculated. Third, the correlation (e.g., Spearman or Pearson’s) is calculated between each single-cell transcriptome and the average transcriptome of each cluster to obtain a given cell’s transcriptome similarity. Fourth, for each cell, the resulting similarities are normalized across clusters of each sample, and the normalized similarities to different samples are concatenated for the final cluster similarity spectrum (CSS) of the cell. The CSS vector representation of each cell is used for downstream clustering, embedding, projection, and other analyses. b We applied CSS to integrate human cerebral organoid data from different individuals, batches, experimental and technical conditions, and technologies