Table 3.

Comparing models created by the supervised clustering algorithms on different samples (ages 18–35 years only).

SVM
	Tel Aviv					Beijing					Cambridge
	Self	GSP	U	P	A	Self	GSP	U	P	A	Self	GSP	U	P	A
z-scores	53	71	51	82	47ns,2	73	75	61	98	71ns,2	66	86	62	80	74ns,ns
PCA	59	72	54	86	59ns,2	74	69	59	95	751,2	79	73	63	93	771,2
DM	72	69	59	98	67ns,2	64	75	57	89	69ns,2	70	78	61	92	762,2
μIDM	73	70	60	97	69ns,2	66	75	58	91	69ns,2	73	78	63	95	782,2
ICPQR	74	64	57	90	64ns,2	63	59	53	96	661,2	77	74	63	97	832,2
ICPQR	54	62	51	93	51ns,2	70	61	54	91	62ns,2	78	71	62	93	761,2
Random forests
	Tel Aviv					Beijing					Cambridge
	Self	GSP	U	P	A	Self	GSP	U	P	A	Self	GSP	U	P	A
z-scores	78	63	58	85	65ns,2	64	67	55	97	792,2	72	81	63	91	832,ns
PCA	70	58	54	88	58ns,2	66	64	55	98	69ns,2	73	77	63	96	76ns,2
DM	70	65	56	92	55ns,2	58	69	53	89	62ns,2	64	81	58	82	69ns,ns
μIDM	72	56	53	84	53ns,2	60	66	53	94	59ns,2	70	73	59	97	70ns,2
ICPQR	70	60	53	86	61ns,2	56	61	51	95	641,2	65	74	57	91	761,1
ICPQR	69	62	54	85	58ns,2	69	65	56	94	66ns,2	71	75	60	96	72ns,2

Self, the percent of brains from the test sample that were correctly classified by a model created on the test sample (applying 10-folds cross-validation); GSP, the percent of brains from the test sample that were correctly classified by a model created on the GSP sample; U, the percent of brains expected to be similarly classified by the GSP and Self models if they were unrelated; P, the percent of brains expected to be similarly classified by the GSP and Self models if they were perfectly overlapping; A, the actual percent of brains similarly classified by the GSP and Self models; the uppercase text marks whether this percentage was significantly different from the percent expected if the models were unrelated and perfectly overlapping, respectively. ns, not significant; 1, p < 0.0014 (alpha following the Dunn–Šidák correction for multiple comparisons); 2, p < 0.0001; PCA, principle component analysis; DM, diffusion mapping, Euclidean distances; μIDM, isometric diffusion map; ICPQR, incomplete pivoted QR decomposition with the kernel of the diffusion map; ICPQRd, incomplete pivoted QR decomposition without the kernel of the diffusion map.