Fig. 1: cyCombine overview.

a Batch correction workflow. First, expression values are transformed in each batch to enable co-clustering of samples from all batches. After clustering, the transformed values are reverted to expression values and ComBat is applied to each self-organizing map (SOM) cluster. b Panel merging workflow. Clustering is performed on overlapping markers, and the missing values for each cell in a panel are imputed using probability draws from the kernel density estimates (kde) from co-clustered cells of the other panel.