Table 3. Comparison of modularity and reproducibility for subtyping methods.

Dataset	Methods	Q	Reproducibility
SMC dataset	Hierarchical Clustering (Euclidian Distance)	—	83.42%
	Hierarchical Clustering (Correlation)	—	86.87%
	Louvain method (Euclidian Distance)	0.0110 (0.1891a)	—b
	Louvain method (Correlation)	0.2202	92.25%
ADNI validation datasetc	Hierarchical Clustering (Euclidian Distance)	—	72.03%
	Hierarchical Clustering (Correlation)	—	89.03%
	Louvain method (Euclidian Distance)	0.0235 (0.1665a)	—b
	Louvain method (Correlation)	0.2464	92.53%

^aTo compute the modularity value, similarity matrix is required but it also affects the modularity value Q. Thus, we computed the value using the same similarity matrix with our method in order to observe the effects of the modular organization only.

^bThe Louvain method with the Euclidian distance raised only two subtypes and thus it is unfair to compare its reproducibility with other methods.

^cThe ADNI validation dataset contained an unknown subtype and we excluded this type in the reproducibility analysis.