RETRACTED ARTICLE: Detangling PPI networks to uncover functionally meaningful clusters

BMC Systems Biology

Table 1 The performance of Louvain run directly on the PPI network versus Louvain plus DSD at different edge removal thresholds; the reported results of Louvain are median values from running the algorithm over 10 random permutations of the nodes. We discard clusters of size < 3

Method	Enriched Clusters	# NEC	% NEC	# NEC S	% NEC S
PPI	29.5/47.5 (62.11%)	799.0	13.10%	548.5	8.99%
4.0	130.0/192.0 (67.71%)	1144.0	18.77%	1011.0	16.58%
4.5	175.0/265.5 (65.91%)	1960.5	32.16%	1562.0	25.62%
5.0	106.5/173.0 (61.56%)	1736.0	28.48%	967.0	15.86%
5.5	15.0/45.5 (32.97%)	361.5	5.93%	288.0	4.72%
6.0	5.0/21.5 (23.26%)	221.0	3.63%	178.5	2.93%

NEC= “Nodes in Enriched Clusters”. We calculate %NEC in two settings: %NEC is enrichment in the GO hierarchy with terms above the fifth level filtered out, and %NEC S uses the same filtered GO hierarchy, but then only gives a node credit if there is a match between one of the node’s labels and one of the terms for which there is GO enrichment for the cluster. Note that without modifying Louvain to restrict the maximum cluster size, the S statistic is the most meaningful. Running directly on the PPI network and run with high DSD thresholds, Louvain produces a relatively small number of clusters, and many are of very large size. It is worth noting that with a DSD threshold of 5, nearly 175 clusters are produced, and the enrichment statistics remain reasonable
Bolded values represent the best values achieved for the %NEC and %NEC S statistics comparing the PPI network and different DSD detangling thresholds

ISSN: 1752-0509