Extended Data Table 5 Benchmarking cell embeddings using scIB and scGraph with default annotations for 9 cell subtypes of fibroblasts, applied to the fibroblast subset of the Human Fetal Lung Cell Atlas. All methods are re-trained on this subset

Method	Bio conservation							Batch correction					scIB aggregate score			scGraph -
Method	I-label	L-NMI	L-ARI	K-NMI	K-ARI	S-label	cLISI	S-batch	iLISI	KBET	G-Con	PCR	Batch	Bio	Total	scGraph -
PCA	0.584	0.707	0.618	0.638	0.581	0.554	0.990	0.910	0.004	0.189	0.952	0.000	0.411	0.667	0.565	0.452
TSNE	0.653	0.510	0.160	0.567	0.445	0.597	1.000	0.641	0.021	0.182	0.748	0.000	0.318	0.562	0.464	0.372
UMAP	0.757	0.542	0.203	0.635	0.490	0.633	1.000	0.663	0.035	0.262	0.904	0.000	0.373	0.609	0.514	0.408
Harmony	0.452	0.232	0.154	0.185	0.101	0.491	0.815	0.863	0.139	0.770	0.820	0.887	0.696	0.347	0.487	0.396
Scanorama	0.542	0.670	0.586	0.535	0.471	0.539	0.994	0.914	0.066	0.517	0.920	0.123	0.508	0.620	0.575	0.352
BBKNN	0.659	0.470	0.178	0.527	0.408	0.580	0.994	0.669	0.076	0.388	0.897	0.000	0.406	0.545	0.489	0.075
scVI	0.493	0.251	0.171	0.265	0.180	0.503	0.811	0.860	0.145	0.677	0.813	0.936	0.686	0.382	0.504	0.339
scANVI	0.541	0.846	0.824	0.663	0.629	0.549	0.998	0.858	0.082	0.701	0.926	0.636	0.640	0.721	0.689	0.356
scPoli	0.569	0.440	0.245	0.475	0.332	0.549	0.951	0.800	0.108	0.623	0.970	0.521	0.604	0.509	0.547	0.273
Geneformer	0.486	0.340	0.149	0.121	0.082	0.504	0.940	0.892	0.081	0.542	0.677	0.558	0.550	0.375	0.445	0.378
scGPT	0.455	0.147	0.046	0.110	0.051	0.473	0.773	0.764	0.161	0.551	0.710	0.589	0.555	0.293	0.398	0.259
scGPT (FT)	0.487	0.338	0.138	0.309	0.195	0.520	0.948	0.739	0.104	0.615	0.921	0.723	0.621	0.419	0.500	0.283
Islander	0.822	1.000	1.000	1.000	1.000	0.803	1.000	0.896	0.073	0.811	0.943	0.000	0.545	0.946	0.786	-0.036
Author’s	0.753	0.569	0.220	0.723	0.615	0.647	1.000	0.683	0.069	0.474	0.946	0.000	0.435	0.647	0.562	0.421

Quick links

Search