| Adjusted Rand Index (ARI) |
 |
 |
[-1, 1] |
1 |
Compares clustering overlap, correcting for random labels. |
| Average Silhouette Width (ASW) |
 |
|
[0, 1] |
1 |
ASW Batch: batch mixing quality; ASW Label: cell-type preservation. |
| Cell Cycle Conservation |
|
|
[0, 1] |
1 |
Measures preservation of cell-cycle signals after integration. |
| Graph Connectivity |
|
|
[0, 1] |
1 |
Measures preservation of cell-type neighborhoods in integrated kNN graphs. |
| HVG Overlap |
|
|
[0, 1] |
1 |
Measures how well gene variability is preserved after integration. |
| iLISI & cLISI |
|
|
– |
The higher the better |
iLISI for batch mixing; cLISI for cell-type preservation. |
| Isolated Label F1 score |
|
|
[0, 1] |
1 |
Measures how well rare or isolated cell types are preserved. |
| kBET |
|
|
[0, 1] |
1 |
Tests if batch labels are evenly mixed in local neighborhoods. |
| Normalized Mutual Information |
|
|
[0, 1] |
1 |
Measures clustering accuracy against known labels. |
| Principal Component Regression |
|
|
[0, 1] |
1 |
Measures how well continuous variation is preserved after integration. |
| Fowlkes-Mallows |
|
|
[0, 1] |
1 |
Compares agreement/disagreement rates between two clusterings. |
| Mutual Information |
|
|
– |
The higher the better |
External metric based on joint entropy. |
| Rand Index |
|
|
[0, 1] |
1 |
Measures consistency between two clusterings. |
| V-Measure |
|
|
– |
The higher the better |
Based on homogeneity and completeness of clustering. |
| Adjusted Mutual Information (AMI) |
|
|
[0, 1] |
1 |
Measures clustering similarity corrected for chance. More robust than NMI when the number or size of clusters varies. |