Table 2.

Benchmark results of ProjectSVR and other reference mapping algorithms.

Task	Method	Accuracy	ARI	Model building (seconds)	Reference mapping (seconds)
PBMC	De novo integration (scVI)	0.791	0.649	NA	NA
	ProjectSVR	0.733	0.585	70.24	21.35
	symphony	0.694	0.577	304.04	3.52
	projectTILs	0.487	0.412	4943.19	1106.34
	Seurat	0.753	0.600	1351.65	677.13
	scArches	0.797	0.689	373.53	156.81
	iNMF	0.640	0.532	100.20	3.53
	SCALEX	0.685	0.527	568.68	99.07
MFI	De novo integration (fastMNN)	0.895	0.848	NA	NA
	ProjectSVR	0.697	0.556	77.07	10.89
	symphony	0.626	0.492	85.97	1.95
	projectTILs	0.107	0.259	1454.18	326.04
	Seurat	0.740	0.577	88.46	862.62
	scArches	0.739	0.553	316.77	110.17
	iNMF	0.566	0.394	74.90	2.54
	SCALEX	0.523	0.401	493.47	61.11
mTCA	De novo integration (scANVI)	0.993	0.995	NA	NA
	ProjectSVR	0.930	0.952	663.98	83.82
	symphony	0.782	0.772	261.56	3.26
	projectTILs	0.690	0.750	4192.72	1081.52
	Seurat	0.916	0.949	960.07	1705.35
	scArches	0.895	0.955	329.19	135.79
	iNMF	0.726	0.671	200.96	5.26
	SCALEX	0.778	0.906	498.32	192.24

Note: ARI: Adjusted rand index, Bold values indicate results from the present method; italic values indicate the best result among all comparison methods.