DiviK: divisive intelligent K-means for hands-free unsupervised clustering in big biological data

BMC Bioinformatics

Table 1 Values of quality indices measured for OSCC data

Clustering algorithm	Global feature engineering method	Adjusted Rand index	Dice index	Relative EXIMS score	Adjusted Rand index rank	Dice index rank	Relative EXIMS score rank	Sum of ranks	Overall quality d(0, 0, 0)	Overall quality d(1, 1, 1)
Spectral	UMAP	0.2792	0.4844	0.5891	17	16	20	53	0.8122	0.9768
Spatial	UMAP	0.0000	0.0000	1.0000	19.5	19	2.5	41	1.0000	1.4142
Spectral	Xception	0.0000	0.0000	1.0000	19.5	19	2.5	41	1.0000	1.4142
K-means	Knee PCA	0.2723	0.0000	1.0000	18	19	2.5	39.5	1.0364	1.2368
K-means	Xception	0.3098	0.4577	0.9197	16	17	8	41	1.0730	0.8814
K-means	EXIMS PCA	0.4827	0.5129	0.8323	12	14	9	35	1.0903	0.7300
Spectral	EXIMS PCA	0.5447	0.7418	0.6449	8	8	18	34	1.1237	0.6325
K-means	none	0.3364	0.5043	0.9712	15	15	6	36	1.1449	0.8288
K-means	UMAP	0.5231	0.7238	0.7225	10	10	13	33	1.1487	0.6170
Spatial	Knee PCA	0.4985	0.7065	0.7639	11	11	10	32	1.1537	0.6272
DiviK	EXIMS PCA	0.6082	0.7765	0.6383	4	5	19	28	1.1749	0.5782
Spectral	none	0.5906	0.7966	0.6520	5	4	17	26	1.1868	0.5745
DiviK	Knee PCA	0.5567	0.7540	0.7289	7	7	12	26	1.1873	0.5749
DiviK	Xception	0.4203	0.6429	0.9395	14	13	7	34	1.2136	0.6835
Spatial	none	0.5617	0.7720	0.7587	6	6	11	23	1.2195	0.5498
DiviK	UMAP	0.6534	0.8369	0.6568	2	3	16	21	1.2485	*0.5143*
Spatial	Xception	0.6517	0.8465	0.6851	3	2	15	20	1.2691	*0.4940*
Spectral	Knee PCA	0.4594	0.6897	0.9891	13	12	5	30	*1.2904*	0.6235
Spatial	EXIMS PCA	0.7035	0.8672	0.6977	1	1	14	16	*1.3167*	*0.4438*
DiviK	none	0.5433	0.7372	1.0000	9	9	2.5	*20.5*	*1.3560*	0.5269

Top three values of a quality index among pairwise combinations of clustering methods and feature engineering procedures are presented in bold italics

ISSN: 1471-2105