12 Batch Effect Assessment

Before proceeding with integration, it is important to evaluate whether technical batch effects — driven by sample of origin or donor identity — are confounding the RNA and ATAC embeddings. If cells from the same sample or donor cluster together regardless of cell type, batch correction is required. If cells from different samples and donors intermix within biologically coherent clusters, the data can proceed to integration without additional correction.

The following plots colour the RNA (ribo-regressed PCA-based) and ATAC (LSI-based) UMAPs by sample, donor, and sample-donor combination. Strong separation by any of these technical covariates is evidence of batch effect.