Table 3.

Overview of computational approaches for prediction of protein complexes from PPI networks. The current state-of-the-art methods are divided into three categories: node-affinity, cluster-quality, and network embedding-based approaches. The input of each method is shown in the second column. A link to the public implementation of each method (if available) along with the year of publication is given in the third column. Other properties such as the number of parameters, the capability of the method to use edge-weights or to predict overlapping protein complexes are given in the last three columns, respectively.

Category	Biological Knowledge/data	Method – Website	Feature(s)
Node Affinity-based approaches	×	MCL [2002]	MCL has 2 parameters and utilizes edge weights. It detects non-overlapping clusters. The size of the clusters depends on the inflation parameter.
		MCODE [2003]	MCODE depends on 5 parameters and does not utilize the edge weights. By setting the fluff parameter, it can detect overlapping clusters. The predicted clusters are of high density. MCODE is unable to find sparse clusters.
		CFinder [2006]	CFinder has 2 parameters and employs edge weights. The predicted clusters have a clique topology. CFinder detects overlapping clusters, while it is unable to find sparse ones.
		AP [2007]	AP has 1 parameter, that affects the cluster formation, and it does not use edge weights. It detects non-overlapping and dense clusters.
		CMC [2009]	CMC has 2 parameters and employs edge weights. The clusters have a clique topology. CMC is unable to find sparse clusters. The size of the clusters depends on the parameters. CMC can detect overlapping clusters.
		PEWCC [2013]	PEWCC has 2 parameters and uses edge weight. It deals with false-positive interactions by introducing a PE-score, while it does not consider the effect of false-negative ones. PEWCC detects highly overlapped and repetitive clusters.
		ProRank + [2014]	ProRank + has 2 parameters and employs edge weights. It considers the effect of false-positive interactions but not the false-negative ones. ProRank + detects overlapping clusters.
		DPC-NADPIN [2016]	DPC-NADPIN has 2 parameters and does not utilize edge weights. It incorporates gene expression data to create a dynamic PPI network. It is unable to predict small clusters. DPC-NADPIN detects overlapping clusters.
		idenPC-MIIP [2020]	idenPC-MIIP has 2 parameters and employs edge weights. It considers the effect of false-positive interactions by calculating MIIP-score. idenPC-MIIP can detect overlapping clusters.
	Microarray data
	DMSP [2007]	DMSP depends on 2 parameters. It considers the effect of false-positive edges by calculating the gene-expression similarity between pairs of protein. DMSP can predict non-overlapping clusters.
Cluster quality-based approaches	×	miPALM [2010]	miPALM has 2 parameters and assigns edge-weights. It detects dense clusters and is unable to predict small and sparse clusters. miPALM predicts overlapping clusters; however, it does not consider the effect of false-positive and false-negative interactions.
		ClusterOne [2012]	ClusterOne has 3 parameters and it utilizes edge weights. It is unable to find small and sparse clusters. ClusterOne predicts overlapping clusters; however, it does not consider the effect of false-negative interactions.
		Core&Peel [2016]	Core&Peel depends on 3 parameters and it uses the edge weights. It predicts dense complexes. The size and density of the clusters depends on 2 parameters. Core&Peel can detect overlapping clusters; however, it does not consider the effect of false-negative interactions.
		IMHRC [2017]	IMHRC has 5 parameters and it employs edge weights. It is unable to find small and sparse clusters. IMHRC can detect overlapping clusters; however, it does not consider the effect of false-negative interactions.
		PC2P [2020]	PC2P is a parameter-free algorithm. It can detect small and large as well as sparse and dense clusters. However, it does not utilize edge weights, but can detects non-overlapping clusters.
		CC [2021]	CC is a parameter-free approach. It can detect small and large as well as sparse and dense clusters. However, it does not utilize edge weights, and can detect non-overlapping clusters.
		OCC [2021]	OCC is a parameter-free approach. It can detect small and large as well as sparse and dense clusters. Although it does not utilize edge weights, it can detect overlapping clusters.
		WCC [2021]	WCC is a parameter-free approach. It can detect small and large as well as sparse and dense clusters. While it utilizes edge weights, it can detect non-overlapping clusters.
		OWCC [2021]	OWCC is a parameter-free approach that uses edge weights. It can detects small and large as well as sparse and dense clusters. OWCC detects overlapping clusters, however it does not consider the effect of false-negative interactions.
		CUBCO [2022]	CUBCO is a parameter-free approach that uses edge weights. It can detect small and large as well as sparse and dense clusters. CUBCO considers the effect of false-negative as well as false-positive interactions; however, it cannot detect overlapping clusters.
	Functional homogeneity	RNSC [2004]	RNSC depends on 7 parameters and it does not consider edge weights. RNSC is a randomized algorithm and in each round, it generates different clusters. It is highly dependent on the initial clusters and it is unable to detect overlapping clusters.
Network embedding-based approaches	×	CPNM [2020]	CPNM has 6 parameters and uses edge weights. It finds non-overlapping clusters. CPNM detects dense clusters and not sparse ones.
		DPCMNE [2021]	DPCMNE is dependent on 5 parameters and uses the edge weights. It is not able to detect sparse clusters, but it can detect overlapping clusters.
	Gene Ontology	GANE [2018]	GANE has 3 parameters and it utilizes edge weights. While it cannot detect sparse clusters, it is able to predict overlapping clusters.