Table 3.

Table summarizing the classification and clustering performance of HYDRA for the experiments using structural MRI and genetic data, respectively. Results are reported for three values of the parameter K. The optimal value of the parameter K that was estimated by performing model selection based on clustering stability is denoted by *. The differences in AUC were statistically insignificant between K = 1 and K = 3 for MRI data (two-tailed t-test p-value equals to 0.115) and between K = 1 and K = 2 for genetic data (two-tailed t-test p-value equals to 0.102). This suggests that discriminative signal was preserved, allowing for clinically relevant clusters to be found.

Experiment		Classification/Clustering Performance
Data	K	AUC	ARI
MRI	1	0.9149 ± 0.0563	—
	2	0.9123 ± 0.0517	0.2054 ± 0.2477
	3*	0.9021 ± 0.0572	0.2724 ± 0.1430
Genotype	1	0.7296 ± 0.1033	—
	2*	0.7047 ± 0.1105	0.7986 ± 0.2266
	3	0.6990 ± 0.1121	0.6412 ± 0.3124