Table 5.
Batch effect correction comparison averaged over 32 simulation settings with balanced batches (500 cells in each batch), excluding settings where batch effect or cell type effects are zero.a
Method | kBET Acceptance | 1 - ASW Batch | ASW Cell Type | LISI Batch | 1 - LISI Cell Type |
---|---|---|---|---|---|
Raw Data | 0.000 | 0.634 | 0.104 | 0.006 | 0.866 |
JIVE | 0.250 | 0.881 | 0.290 | 0.323 | 0.935 |
Seurat | 0.317 | 0.976 | 0.158 | 0.462 | 0.878 |
Harmony | 0.708 | 0.999 | 0.142 | 0.790 | 0.880 |
Liger | 0.428 | 0.922 | 0.158 | 0.449 | 0.925 |
Combat-seq | 0.104 | 1.000 | –0.018 | 0.400 | 0.682 |
Higher values indicate better mixing of different batches or better retention of cell type effects. The highest average value for each metric is in bold. kBET results are based on a 5% sample size; results were similar at larger sample sizes.