TABLE 1.

Overview of different clustering algorithms^a

Algorithm	Cluster no. criterion	Unassigned nodes	No edge filtering required	Preferred network type	No parameter tuning
manta	Optimize sparsity	Yes	Yes	Undirected	Yes
WGCNA signed	Dynamic branch cut	Yes	Yes	None (constructs network)	Yes
WGCNA unsigned	Dynamic branch cut	Yes	No	None (constructs network)	Yes
Louvain method	Optimize modularity	No	Yes	Undirected; directed version possible	No
MCL	Parameter dependent	No	Yes, if parameters optimized	Undirected	No
Girvan-Newman algorithm	User dependent	No	No	Undirected; directed version possible	Yes
Kernighan-Lin bisection	Only bisection	No	Yes	Any	Yes

^aDifferent properties of manta, WGCNA, MCL, Louvain community detection, the Girvan-Newman algorithm, and the Kernighan-Lin bisection algorithm. The following properties are summarized: how algorithms choose a cluster number, whether they can leave nodes unassigned, whether they perform better with negatively weighted edges removed, and what types of networks they accept. Finally, we assessed whether algorithms required extensive parameter tuning before achieving optimal performance on simulated data.