2.5 Co-occurrence network analysis
Network analysis was performed on the rarefied OTU table and only OTUs with relative abundances over 0.1% and presented in at least 3 samples were included. The co-occurrence network analysis was executed in the CoNet 3 plugin for Cytoscape 3.7.1. For each network, node attributes were computed by bootstrapping approach with 1,000 iterations. Cluster Viz, a plugin in Cytoscape was used to find clusters in a network. Network topological parameters were calculated by the NetworkerAnalyzertool in Cytoscape 3.7.1. Networks were constructed based on five correlation measures using the Brown method, and only correlations with a ρ value over 0.8 and a significance level below 0.06 were presented in the network. In a network, high average node degree, high clustering coefficient and high network density can depict how dense a network is and the tendency to form clusters (Banerjee et al., 2019; Rebollar et al., 2019). In a microbial co-occurrence network, the OTUs with the highest degree, highest closeness centrality and lowest study, OTU with degree greater than 10, closeness centrality greater than 0.24, and betweenness centrality lower than 0.011 was selected as the keystone taxa.